Documente Academic
Documente Profesional
Documente Cultură
ORIGINS
Studies in the sources of scientific creativity
Volume 2
Managing Editor:
Fernand Hallyn, University of Ghent, Belgium
Editorial Board:
Gillian Beer, Cambridge University, U.K.
James J. Bono, State University of New York, Buffalo, U.S.A.
Marc de Mey, University of Ghent, Belgium
Thomas Da Costa Kaufman, Princeton University, U.S.A.
Peter Galison, Harvard University, U.S.A.
Paolo Galluzzi, Istituto e Museo di Storia delle Science, Firenze, Italy
Rom Harre, Oxford University, U.K.
Peter Machamer, University of Pittsburgh, U.S.A.
Arthur I. Miller, University College London, U.K.
William Shea, University of Strasbourg, France
Gerard Simon, University of Lille III, France
Geert Vanpaemel, University of Leuven, Belgium
Peter Weingart, University of Bielefeld, Germany
SCOPE
The aim of the series is to present historical and theoretical studies on the sources of
scientific creativity. The series provides a platform for various transdisciplinary
viewpoints.
Indeed, on the one hand, the origins of scientific creativity should be studied in the
light of its relations with sources of creativity in other disciplines (literary, artistic),
in order to illuminate the particular scientific element in the genesis of scientific in-
novation.
On the other hand, the complexity of the topic necessitates a variety of approaches,
where logic, cognitive studies, poetics, rhetoric, history of ideas and other disciplines
meet in a common interrogation.
In short, the series welcomes studies which integrate philosophy and history of
science in a broad, diversified field of research, where there is room for a great va-
riety of perspectives with different methodological and conceptual references and
where isolationism as well as reductionism are avoided.
Inconsistency in Science
Edited by
JOKEMEHEUS
Ghent University, Ghent, Belgium
....
"
Springer-Science+Business Media, B.V.
A C.I.P. Catalogue record for this book is available from the Library of Congress.
Preface Vll
Approximate Truth 81
Bryson Brown
v
VI CONTENTS
INDEX 213
PREFACE
Within traditional philosophy of science, the role of inconsistencies has largely been
ignored. At best, inconsistencies were seen as a hindrance for good scientific
reasoning. The reason for this is not difficult to understand. Until very recently,
rationality has been identified with reasoning according to Classical Logic. And, as
is well known, Classical Logic does not allow one to reason sensibly in the presence
of inconsistencies.
Today, it is generally recognised that almost all scientific theories at some point
in their development were either internally inconsistent or incompatible with other
accepted findings (empirical or theoretical). A growing number of scholars
moreover recognises that inconsistencies need not be disastrous for good reasoning.
Three developments were central for this change in attitude. First, there is the
shift in attention from finished products to discovery processes. Whereas finished
theories usually seem to satisfy the consistency requirement, developing theories
typically do not. Next, there is the availability of case studies on episodes in the
history of the sciences that involved inconsistencies. Finally, there is the study of
paraconsistent logics' that started some fifty years ago. This study not only
challenged the view that Classical Logic would be the appropriate tool for reasoning
in all contexts, but also resulted in a variety of systems that are adequate for
reasoning in inconsistent contexts.
The prevalence of inconsistencies in the sciences raises a large number of
interesting questions. How should 'logical anarchy' be avoided? Should one resort
to a paraconsistent logic or to non-logical criteria? What about the acceptance of
inconsistent theories? Are we ever justified to accept an inconsistent theory, and if
so, can this acceptance be more than provisional? How does inconsistency affect our
notion of truth? Should inconsistent theories at best be considered as 'containing
some truth' or can they be considered as true in the same strong sense as consistent
theories? Do inconsistencies exist in the world out there or only in the theories we
, A logic is called paraconsistent iffit does not validate Ex Fa/so Quodlibet (A, -A f-- B).
Vll
Vlll PREFACE
humans design about that world? The obvious importance of these questions
certainly warrants to devote a series of studies to the theme.
The incentive for this book was the First World Congress on Paraconsistency
(Ghent University, Belgium, 30 July - 2 August 1997). At this congress, a workshop
on The Role of Inconsistencies in the History and Philosophy of Science was
organized. This was not the first meeting devoted to the theme. However, the Ghent
workshop was at least for two reasons unique. Never before had a meeting on this
subject been attended by specialists from all over the world. And even more
importantly, never before had philosophers of science and historians of science met
with logicians and computer scientists to discuss this intriguing theme. A selection
of papers presented at this workshop is included in the present volume. In order to
do justice to the variety of approaches, this selection has been extended with a
number of invited papers.
The book opens with two contributions from the philosophy of science, "From
Copernicus to Ptolemy: Inconsistency and Method" by Thomas Nickles and
"Inconsistent Reasoning toward Consistent Theories" by Arthur Miller. Nickles
compares the standard theory-centred conception of science with newer pragmatic
and model-based accounts of scientific inquiry regarding their methodological
treatment of inconsistencies. Miller investigates different sources of inconsistencies,
and draws some conclusions from this with respect to scientific progress. These are
followed by two studies in the philosophy of mathematics, "Inconsistencies in the
history of mathematics: the case of infinitesimals" by Jean Paul Van Bendegem and
"Mathematical Change and Inconsistency: A Partial Structures Approach" by Otavio
Bueno. Van Bendegem explores an alternative approach for limit analysis that
solves the inconsistencies connected with infinitesimals; the case is further used to
defend a contingent view on mathematical progress. Starting from da Costa's and
French's partial structures approach, Bueno presents a framework for mathematical
change that assigns a positive role to inconsistencies. He applies the framework to
the development of set theory.
Next, there are several analyses of inconsistencies in the empirical sciences that
are based on a specific paraconsistent approach, "Approximate Truth: A
Paraconsistent Account" by Bryson Brown, "Inconsistency in Science: A Partial
Perspective" by Newton da Costa and Steven French, "Inconsistency and the
Empirical Sciences" by Graham Priest, "In Defence of a Programme for Handling
Inconsistencies" by Diderik Batens, "How to Reason Sensibly yet Naturally from
Inconsistencies" by Joke Meheus and "Why the Logic of Explanation is
Inconsistency-adaptive" by Erik Weber and Kristof De Clercq. The first three of
these defend each a different realistic view on inconsistent theories. Brown presents
an account of approximate truth that is based on a specific non-adjunctive approach
to paraconsistency, and that aims at explaining the success of past and current
(possibly inconsistent) theories. da Costa and French offer a model-theoretic account
in which theories are regarded as partial structures. On this account, inconsistent
theories (just like consistent ones) can be considered as partially true and accepted as
such. Priest advocates the view that reality itself is inconsistent and that
contradictions are observable. In line with this, he defends the idea that inconsistent
theories can be considered as true in the strong sense of the word. The last three
PREFACE IX
papers in this group explore the philosophical foundations and some applications of
inconsistency-adaptive logics. Batens spells out the philosophical programme
underlying this family of logics; one of his central arguments is that handling
inconsistent theories requires a logic that stays as close as possible to Classical
Logic. Meheus argues that reasoning from inconsistencies requires an extremely rich
inconsistency-adaptive logic, and presents a system that meets this requirement.
Weber and De Clercq argue that inconsistency-adaptive logics are much better
suited than Classical Logic to define the different types of explanation.
The volume closes with two case studies, "A Paradox in Newtonian Gravitation
theory II" by John Norton and "Inconsistency, Generic Modeling, and Conceptual
Change in Science" by Nancy Nersessian. Norton provides a rigorous yet
transparent demonstration of the inconsistency of Newtonian Cosmology, and
defends the view that physical theorists handled this inconsistency by a content
driven approach rather than a logic driven approach. Nersessian analyses Maxwell's
construction of the laws of electrodynamics, and shows how generic modeling
enabled him to tolerate several inconsistencies in his derivation.
Acknowledgments
The workshop The Role of Inconsistencies in the History and Philosophy of Science
was organized by the international research community "Science and Culture" and
was sponsored by the Fund of Scientific Research - Flanders. The editor is a
Postdoctoral Fellow of the same Fund.
The editor wants to thank the members of the organising committee of the First
World Congress on Paraconsistency for their help in organising the workshop and
Isabel D'Hanis for her help in preparing the manuscript.
Joke Meheus
THOMAS NICKLES
Abstract: In recent years several philosophers and other science studies experts have adopted a somewhat
more Ptolemaic than Copernican view of theories, models, and scientific research, namely, the "semantic"
conception of theories and their applications. On the old, "Copernican" view, theories are deductively
integrated, interpreted logical calculi, in standard symbolic logic, and science is a theory-centered
enterprise that aims for a comprehensive, single picture of the universe. Accordingly, consistency
becomes a hard constraint, a sine qua non of useful, rational inquiry. Proponents of the semantic
conception are somewhat more "Ptolemaic" in treating theories as collections of models and in placing
the solution of local problems in restricted domains ahead of grand, spectator theories of the universe.
They focus as much on the process of inquiry as upon the logical structure of its products. Remarkably, in
scientific practice we find inconsistencies and near-inconsistencies of various kinds apparently popping
up everywhere, suggesting that one cannot lay a tight grid of standard logic over scientific practice.
However, deflating unitary theories into collections of models makes inconsistency harder to define and
locate precisely-but also less serious. Consistency often becomes a soft, negotiable constraint--one
constraint among others. In practice, inconsistency rarely leads to complete disaster, for there are various
ways to tame it. I urge paraconsistency and inconsistency-tolerant logicians to pay attention to real
scientific examples of scientists' responses to inconsistency and quasi-inconsistency, e.g., conceptual
incongruities such as conceptual "blowups" and pragmatic inconsistencies such as the simultaneous use
of mutually inconsistent models.
1. INTRODUCTION
* Acknowledgement. I thank the U.S. National Science Foundation for support on projects concerning
heuristic appraisal and knowledge pollution. Thanks also to Joke Meheus for helpful suggestions. She
is not responsible for infelicities that remain.
I This is not an absolute dichotomy, of course, since nearly everyone regards theories as problem
solutions of a sort. However, problem-solving accounts of research tend to convey a quite different
picture than do theory-centered accounts. The difference is philosophically, and logically, important.
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 3
other words, their stance is not consistent even at the metal eve I, insofar as
'consistent' implies fixed. Although this makes the work of science studies more
difficult, it (again) does not seem to produce intellectual disaster. On these issues,
from time to time we probably all feel strongly both ways!
So reads a fragment of Copernicus's famous letter to Pope Paul III that prefaces his
De Revolutionibus (quoted from Kuhn 1959, 138 f).
The inconsistencies that Copernicus complained of were of two related kinds.
First, different Ptolemaic mathematicians employed distinct methods and
constructions to explain the same phenomena, constructions that are mutually
incompatible if realistically interpreted, e.g. an equant versus more epicycles.
Second, the same mathematician could use incompatible constructions to explain
different phenomena, or different aspects of the same phenomenon or object. For
example, the constructions used to predict the size of the moon at a given time were
incompatible with those used to predict its location or its speed. Ptolemaic
astronomy made no demand that these methods of calculation should be compatible.
Now even though many phenomena could be predicted as accurately as you
please (given the limits of observational accuracy) by means of some construction or
other, this sort of empirical adequacy was not enough for Copernicus. For him,
construction methods were not merely tools for prediction; rather, they generated
representations of nature that we should be able to piece together into one consistent
and complete picture of the cosmos. Copernicus took it as a given that nature itself is
consistent and had been made so by a rational and aesthetically sensitive God.
I pondered long upon this mathematical uncertainty in establishing the motions of the
system of the spheres. At last I began to chafe that philosophers could by no means
agree on anyone certain theory of the mechanism of the Universe, wrought for us by a
supremely good and orderly Creator. ...
As Kuhn notes, Copernicus here put his finger on the most striking difference
between his and the Ptolemaic theories. The latter permitted one to expand and
shrink the orbit of a given body at will, leaving the others fixed. Copernicus and his
followers believed that a realist interpretation would help to solve both problems
(inconsistency and incompleteness) by imposing available but hitherto ignored
constraints on astronomical problem solving.
From this point in Copernicus's preface, we may draw a direct line to contempo-
rary physicist Stephen Weinberg, writing in Dreams of a Final Theory (1994):
Our present theories are of only limited validity, still tentative and incomplete. But
behind them now and then we catch glimpses of a final theory, one that would be of
unlimited validity and entirely satisfying in its completeness and consistency .... [po 6]
[In addition to multiple symmetries, a final theory would manifest] the beauty of
simplicity and inevitability-the beauty of perfect structure, the beauty of everything
2 Lloyd (1978) rejects the widespread view that Ptolemy himself and the commentators cited by Duhem
(1969) were instrumentalists rather than realists. Laudan (1983, §2) notes that many astronomers had
abandoned Aristotelian demonstrative science with its aim of deducing causal explanations of
celestial motion from self-evident first principles. (I) They gave up attempting to provide causal
explanations in terms of essences, and (2) they adopted a hypothetico-deductive method of testing
their models. Thus Copernicus and Kepler were conservative revolutionaries in trying to return
astronomy to the status of a genuine science from that of a calculative craft.
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 5
fitting together, of nothing being changeable, of logical rigidity. It is a beauty that is
spare and classic, the sort that we find in the Greek tragedies. [po 149]
Other authors, and Kuhn himself in other passages, have been less confident of the
superior accuracy of Copernican theory. Thanks initially to Kuhn's book, we now
appreciate that, at that time, the Copernican theory did not obviously possess all the
advantages often claimed for it (simplicity, predictive accuracy, elimination of
epicycles, etcV For example, Copernicus still retained epicycles, although he
eliminated Ptolemy's five major epicycles as well as the equant (Kuhn 1959,
chap. 2). Nor was Ptolemaic astronomy in crisis, although, in the later Structure of
Scientific Revolutions (1962), Kuhn insisted that every revolution must be preceded
by a crisis. Moreover, in the "Postscript" to the second edition (1970) of Structure,
Kuhn himself devalued overarching theories based on principles such as F = rna, in
favor of concrete problem solutions or "exemplars". Accordingly, it is no longer
clear whether we should count Copernicus's achievement as an exemplar (a
paradigm in Kuhn's "small" sense) or as the creation of a disciplinary matrix (a
paradigm in the "large" sense). While it conveyed a revolutionary new worldview,
only much later was it parlayed into a new disciplinary matrix. Be that as it may, the
Copernican Revolution remained paradigmatic of a scientific revolution in Kuhn's
own development. 4
The word 'epicycle' has become pejorative, signifying a degenerating program
that saves itself through otherwise arbitrary, ad hoc patches. Yet, even here,
Copernicus's and Kuhn's complaints about the Ptolemaic tradition can be
challenged. For the Ptolemaic practice of introducing new epicycles (or any other
constructions that might be convenient) in order to provide a tighter model of a
phenomenon is not completely out of keeping with current scientific practice
(Nickles 1987).
Of course, the Copernican debate is not my present concern. I do not here
question the usual story of scientific progress from Copernicus to Newton's theory
with its integration of terrestrial and celestial mechanics. But I do want to raise the
question whether the Copernican paradigm, including Copernicus's demand for total
consistency, should be taken as a model for all of science.
The subject of his first book-the Copernican Revolution-became the paradigm for
Kuhn's own conception of scientific revolutions: the Copernican theory (if not
Aristotle's theory of motion) was his paradigm of paradigms. On the most common
reading of Kuhn (for which Structure provides ample support), scientists strive to
construct comprehensive, consistent, true representations of the world, unified and
preferably aesthetically pleasing world pictures, although they in fact achieve only
historically limited perspectives. Hence the title of Kuhn's infamous Section X:
"Scientific Revolutions as Changes in World View", and hence the early Kuhn's
love of the Gestalt-switch metaphor. A Kuhnian paradigm is so highly integrated
that it must change holistically.
However, as Joseph Rouse (1987, 1996) has pointed out most clearly, Structure
develops a second account of science that does not sit comfortably with the first,
namely a more Baconian account that emphasizes scientific practice and routinely
producible results over theory. So depicted, scientific work is more pragmatic and
opportunistic than representational; it aims to solve puzzles and to manipulate nature
more than to construct world pictures that copy nature. It aims more at doing and
making than representing. It is as if the big pictures are largely rhetorical by-
products, perhaps even advertisements, that catch the eye of the general public,
funding agencies, and realist philosophers who are uninterested in the practical and
technical details that are the meat and potatoes of real science.
In Kuhn's "Postscript", theory practically disappears in favor of exemplary
problem solutions (small, microparadigms) on the one hand and socio-disciplinary
matrices (macroparadigms) on the other. Also, the theories or macroparadigms that
he mentions there are not unified, integrated, logical-deductive systems at all. For
example, F = rna becomes a schema or template rather than a general law of nature
from which specific force laws are logically deducible. Kuhn} is theory-centered,
while Kuhn2 is practice-centered and theory de-centered.
Kuhn's double account already gives us reason to doubt whether we should
project the Copernican paradigm on all of science. A distind but related reason for
doubt highlights the major features of Baconian science-the emphasis on
experimental intervention. Kuhn himself distinguished the experimental from the
mathematical tradition in the history of science, although others have gone much
further in studying experimental practices. Broadly speaking, Baconian science blurs
the theory-practice distinction and falls within the "maker's knowledge" tradition
according to which (in Vico's later, memorable phrase) we understand best that
which we ourselves have made. On this view, the primary products of scientific
research are practices or techniques or recipes for getting specific results from the
manipulation of nature-tools that can be used in turn in future investigations and
applications. A Baconian demonstration, unlike a geometrical demonstration,
5 Distinct because one can also speak about practice vs. high theory in the Copernican tradition.
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 7
establishes that we can reliably produce some desired effect without necessarily
pretending to have the representational truth about the universe. 6
By contrast, the Copernican world picture provides "spectator's knowledge", the
knowledge characteristic of our contemplation of an aesthetic object rather than
knowledge gained by making or doing. For we cannot experimentally manipulate
the planets and stars; we can only describe their observed motions. Spectator's
knowledge tames nature by aestheticizing it rather than by controlling it, and it
confers a misleading plausibility upon simple, passive, observational empiricism.
'Theory' in this sense retains the connotations of aesthetic contemplation present in
its Greek root, theorein. From this point of view, it is not surprising that Bacon did
not take Copernicus as a model.
Although the maker's knowledge tradition extends back to Bacon and beyond, it
was the American pragmatists-Charles Peirce and (especially) William James and
John Dewey-who drew the distinction in terms of "spectator's" versus "maker's"
and "doer's" knowledge, in the course of rejecting the traditional, Greek distinction
between theory and practice. 7 Wrote Dewey (1908, §V):
Sub specie aeternitatis? or sub specie generationis? I am susceptible to the esthetic
charm of the former ideal-who is not? There are moments of relaxation: there are
moments when the demand for peace, to be let alone and relieved from the continual
claim of the world in which we live that we be up and doing something about it, seems
irresistible; when the responsibilities imposed by living in a moving universe seem
intolerable. [Dewey's emphasis]
6 See Kuhn 1977, chap. 3 ("Mathematical Versus Experimental Traditions in the Development of
Physical Science") and chap. 5 ("History of Science"). For Bacon and maker's knowledge, as well as
a critique of Kuhn, see Perez-Ramos 1988, esp. chap. 3. Mathematical knowledge, too, could be
considered maker's knowledge of an abstract kind. In fact, it was a favorite example, from Hobbes to
Vico and beyond. We understand mathematics so well because we made it. Writers from the mid-
19th-century on increasingly distinguished pure mathematics from empirically interpreted
mathematical assertions.
7 Aristotle had distinguished genuine scientific knowledge from both the knowledge of practical action
and the craftsman's knowledge of making. And, of course, Marx had preceded the pragmatists in
rejecting the theory-practice distinction.
S For simplicity, I am conflating several disputes that might be distinguished in a more expansive
treatment. Some of the differences will fall out below. E.g., many of the positivists were theory-
centered but not raving realist-representationalists.
8 THOMAS NICKLES
The historical Copernicans, including Kepler, Galileo, and Newton, initiated both of
the primary, modem strategies for promoting the consistency of scientific work. The
first is that all the knowledge in a domain be formulatable systematically within a
single, consistent, deductive mathematical theory framework. The second is that all
the major calculative devices employed be realistically interpreted. Since nature
itself is supposedly consistent (whatever, exactly, that could mean), realism imposes
an additional, consistency constraint beyond the usual constraint of empirical
adequacy. The two approaches combine in the idea that the aim of any given science
is to produce a general, deductive theory of its domain that is semantically as well as
syntactically consistent, because it possesses a true, realist interpretation, i.e., a
model in the formal sense. Of course, realism does not preclude disagreement
(inconsistency) among competing scientists and their theories over what is real, nor
does it guarantee that a theory developed in one domain will be consistent with a
theory developed in another domain. However, realism does require that all such
inconsistencies be resolved in the long run.
What is a scientific theory? The main positivist answer9 was that a theory, in
rationally reconstructed form, is a partially interpreted logical calculus. A theory is a
deductive axiom system plus an intended model that corresponds to some domain of
empirical reality (or rather a partial model, a model for the observational statements
of the theory). Insofar as finding good theories is the central aim of science, the
definition of the consistency of theories, and of the body of scientific claims as a
whole, is straightforward. An axiom system is consistent in a proof theoretic sense if
9 Otto Neurath, for one, held a very different view. See Cartwright et al. 1996.
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 9
(to use one common definition) P and not-P are not both provable in the system. It is
consistent in a semantic sense if it has a model, an interpretation that makes the
axioms true. A science is consistent if its central theory or theories are internally
consistent, mutually consistent, and also consistent with the accepted empirical
phenomena and with principles such as conservation of energy. Of course, definition
is one thing, practical determination another. The formal language necessary to
express such a science will be too rich to admit of internal consistency proofs.
Inconsistency in the semantic sense can be a bit tricky to define, since only the
observational terms of the theory and its correspondence rules are interpreted
directly. The theoretical terms are left uninterpreted. But this problem can perhaps
be handled by extending or relativizing the partial models. However, there is no way
to guarantee the future consistency of the theory with the phenomena, given
problems of induction. On the standard view of Popper and others, such
inconsistencies normally mean that we must reject at least one of the theoretical
premises in the (allegedly) deductive derivation of the test prediction. Kuhn demurs
at this point.
Internal inconsistency of a theory was anathema to the positivists-the mark of
irrationality, the complete breakdown of logic and reason. For in standard logic and
epistemology, as everyone knows, a single inconsistency is logically and
epistemically disastrous: an inconsistency anywhere instantly propagates to generate
inconsistency everywhere. In other words, theories, on the positivist-Popperian
conception, are logically brittle. Accordingly, for the positivists, and for most
traditional logicians and philosophers, consistency was not one constraint on rational
inquiry among others, to be balanced in cost-benefit tradeoffs with those others.
Rather, it was an absolute, sine qua non for rational inquiry. For them an
inconsistent system was practically worthless.
In recent decades a quite different conception of theories-the so-called
semantic view or non-statement view of theories-has become popular. There are
many versions of the semantic view. One stems from Evert Beth, who influenced
both Bas van Fraassen and Frederick Suppe, another from Patrick Suppes and his
student Joseph Sneed, who influenced Wolfgang Stegmilller, who in tum inspired
Ulises Moulines, Wolfgang Balzer, and many others, especially in Europe.
Stegmilller dubbed his line the structuralist theory of theories. Thomas Kuhn is
another direct inspiration to the Sneed-Stegmilller line and also to Ronald Giere. I
also mention Nancy Cartwright, whose "Stanford School" approach to theories and
models owes much to Suppes, although she declines the label "semantic view".
What is common to these quite different lO versions of the semantic conception is,
first, the move of identifying a theory directly with a mathematical structure or set of
structures, especially a model or set of models in the semantic sense (more or less),
rather than with a linguistic entity of any kind. Since a syntactic calculus is
linguistic, the non-statement view is very different from the traditional conception of
theories as a cluster of related law-statements axiomatizable in some formal
10 F. Suppe's prologue to his 1989 gives his personal account of the various approaches. For a survey of
the structuralist approach of Sneed, StegmOller, e/ al., see StegmOller 1976 and Balzer e/ al. 1987. See
van Fraassen 1987 and Giere 2000 for their perspectives.
10 THOMAS NICKLES
language. A second common feature is that the semantic theorists (as we may call
them) blur the distinction between a theory and its applications. Its primary
applications are part of the theory itself. As already hinted, to a first approximation,
we can identify a theory with its set of intended models, or with a core plus the
models (e.g., Moulines 1996).
Third, most semantic theorists regard robust theories as a flexible collection or
family of models that are integrated far more loosely than the deductive applications
of a logical calculus. For example, Stegmiiller and company speak of the intended
models of a theory as being related by Wittgensteinean family resemblance rather
than by necessary and sufficient logical conditions. Typically, the formulators of a
theory will directly specify a set of exemplary models that the theory is intended and
expected to handle, and this set will indirectly specify other, future models
(applications of the theory), namely those that sufficiently resemble the former. I I
Each model is a specialization of some part of the theory core, but this cannot be
understood as a deductive relation.
Since similarity is a rhetorical trope, and since exemplary applications function
much as rhetorical topoi or "places", we can say that theory structure, on the
semantic view, turns out to be both more rhetorical and less logical than it was
previously thought to be. 12
A fourth commonality is the denial that theories and their laws are universal in
scope. Rather, their domain is restricted to the open set of their intended
applications. Stegmiiller goes so far as to say that theories ultimately determine their
own scope in a pragmatic manner: those models tum out to belong to a theory that
the theory actually succeeds in incorporating. Repeated failure (as in Kuhnian
normal science) can lead simply to the denial that that model-domain is within the
scope of the theory after all, rather than to refutation of the theory. Thus an
"inconsistency" is resolved partly by decision.
Taking these commonalities together, we can say that the semantic theorists, to a
greater or lesser degree, advance a pragmatic, deflationary view of theories.
Theories are brought down to earth and related directly to human practices and
(especially in the case of Giere) to human cognition. Models and their intended
applications are where the action is. In short, the old view put the cart before the
horse. Syntax is not the really important thing but only a vehicle for conveying the
semantic and pragmatic message. Proponents rightly claim many additional
advantages for the semantic conception. 13
This profound change in the conception of theories and their relation to
applications raises several questions about what inconsistency is and how it is to be
identified and dealt with in practice. Among them, surely, are these. (1) What counts
as an internal inconsistency, given that a theory's models are related by similarity
and not fully deductively integrated? (2) If models are models in the semantic sense,
how can one and the same theory have multiple models?14 (3) What counts as an
inconsistency between a theory and a phenomenon (or "the evidence")? (4) How
determine inconsistency between "theory" (as either held or stated) and practice? (5)
What counts as a mutual inconsistency in either the views or the practices of two or
more researchers? (6) What additional twist is given to the above questions given
that models are almost always idealized and applied to the real world only to an
approximation?
These questions all arise within the relatively formal approach of the structur-
alists (e.g., Stegmiiller 1976, 13 and later) and all the more within the looser
formulations of Cartwright and (especially) Giere. A couple of the questions have
relatively easy answers, but most do not. But neither does inconsistency seem to be
such a desperate problem as it was for those positivists and others for whom
inconsistency, like paradox, demanded a fully rigorous, deductive axiomatic
response.
Not coincidentally, all of the above questions arise already for Kuhn's view of
theories, according to which theories lack a tight deductive structure, and theoretical
knowledge and practice consists of knowing how to employ a set of exemplars
(exemplary problem solutions or "models" in the structuralist sense). If David
Hilbert (1899) found small gaps in that traditional exemplar of a deductive system-
Euclidean geometry-then, on the Kuhnian proto-semantic view, a standard
axiomatization of, say, classical mechanics, is a veritable sieve! Kuhn also denied
that paradigmatic theories are falsified in the traditional (modus tollens) manner
(e.g., the blame for failure is often shifted away from the theory and onto the
practitioner-an extreme extension from logic to practice of Duhem's point about
the complexity of scientific testing). Third, Kuhn pointed out two other kinds of
apparent contradictions that do not seriously disrupt normal scientific activity. One
is that we find quite a bit of slack in the way textbooks present theories. At the
metal eve I mutual inconsistencies are frequent in the sense that one text's definition
is another's fundamental law. Yet inconsistencies at this apparently foundational
level matter surprisingly little to scientific practice. 15 The same can be said for
13 Stegmiiller (1976, chap. I), following Sneed, already provides a long list of advantages of the
structuralist theory, including several that better interpret and justify Kuhn's claims about normal and
revolutionary science.
14 A logical calculus may have any number of interpretations of course, including one in the natural
numbers. But each interpretation corresponds to a distinct theory. No one says that Maxwell's
electromagnetic theory is about the natural numbers.
15 Quine's "Two Dogmas of Empiricism" (1951) rubs out the positivist distinction between conventional
definitions and content-laden axioms or postulates, as a distinction without a difference. Hence, this
12 THOMAS NICKLES
the static or "temporal cut" sense from a theory in the dynamical or evolutionary
sense, and also distinguish models as interpretations of axiomatic formulations
already in hand from phenomenal models in search of suitable axiomatic
formulations.
Roughly speaking, a theory is true of those model-domains that it handles
adequately and false of those that it does not; but this is misleading, because no
theory claims to cover everything. Moulines (1996, 6) writes (contra the Weinberg-
type position quoted earlier): "There is no such thing as a theory of everything, nor
is it reasonable to expect that such a theory will ever show up." Following Kuhn and
Stegmuller: if a theory fails to handle a given model, we can decide to keep the
theory (on the basis of the utility of its previous successes) and simply deny that its
scope extends to that model. This removes the apparent clash of the theory with the
phenomena of this domain. Do not ask of a theory or model, "Is it true or false?" but
"Where does it apply?" A scope decision can make an apparent inconsistency
vanish.
I now tum to Giere and then to Cartwright. Giere's approach is still less formal
and more pragmatic than that of the structuralists. While axiomatization has a
reduced status for the structuralists, it has virtually no role in Giere's account. He is
even more anxious than they are to connect theories with human practice in the
sense that he wants the models of a theory to work also as models in the cognitive
psychological sense of "mental models".
In Understanding Scientific Reasoning (1979, 1984), Giere conceives the real
world as made up of natural systems that interact with one another. A theory
consists of two components: at least one theoretical model plus at least one
theoretical hypothesis linking the ideal model to the world. A theoretical model in
Giere's sense is a human construction that characterizes an idealized system. It is
given by an explicit definition that stipulates how the various terms are to be used.
For example, Newton
defined a theoretical model corresponding to what we now call Newtonian Particle
Systems .... A NEWTONIAN PARTICLE SYSTEM is a system that satisfies the three
laws of motion and the law of universal gravitation. [1984, 8lf; Giere's emphasis]
The models themselves are not an interpretation of anything, although clearly they
are often constructed with that ultimate intention. Theoretical hypotheses employ
theoretical models to make contingent statements about the world, such as the claim
that the solar system is a Newtonian particle system. Here is where interpretation
comes in. A theory is a conjunction of all of the local claims and applications of the
corresponding model(s). Thus Newton's theory is the conjunction of claims about
the solar system, Jupiter and its moons, double stars, cannonballs and baseballs on
the earth, moon, and Mars, etc., etc., as Newtonian particle systems. Giere quickly
adds that most scientific theoretical claims are approximations claiming only
sufficiently interesting similarityl? of structure (his term) with natural systems:
I? In later writings, Giere prefers to speak of better fit rather than similarity.
14 THOMAS NICKLES
In his later book, Explaining Science (1988), the Kuhnian flavor of Giere's view
becomes more explicit. One of Kuhn's central points was that scientists resort to
methodological rules only when normal research breaks down. In normal science
they employ exemplars-paradigm cases of problems-plus-solutions that point the
way in research and that are incorporated in leading textbooks. Every student of
elementary mechanics learns to solve the problems of falling bodies and projectile
motion, the two-body problem, problems of simple harmonic motion, and so on.
Kuhn emphasized that learning these exemplary solved problems, in the strong sense
of learning how to apply them to the standard problems given at the end of the
textbook chapters, is more important than internalizing the symbolic formulas as
descriptions (representations) of reality. Practical knowledge is primary. Students
who fail to develop the practical knowledge of how to solve the problems really
have not learned physics (or whatever) and have, at most, gained a superficial,
aesthetic understanding of what the theory says, a kind of philosophers' gloss on the
theory. As Giere describes his position in Explaining Science,
I argue that theories are typically presented in scientific textbooks as families of models.
I speculate that textbook presentations might reflect the general way humans perform
tasks of cognitive organization, as exemplified in research on chess masters and on
problem solving by trained physicists. [po 278]
For Giere (1988, 86), as for the pragmatists, the various actual and possible
scientific theories form such a diverse domain that there is no demand for a general
"theory of theories". (By comparison, the structuralists do see a structural
commonality worth articulating.) Theories possess no essential logical structure in
common, and, individually, theories need possess no tight, logical unity.
As for Kuhn on exemplars, the important units of and for analysis are smaller
than theories-namely, the models and their applications. A theory is just a
convenient grouping of sufficiently similar models and their principal applications. 18
Furthermore, Giere agrees with Kuhn that the centrality of models to scientific
practice is cognitively significant. This is not cognitive significance in the old,
positivist sense but in the sense of the cognitive sciences and sociology.19 Theories
and models are not simply candidate representations of nature, nor are they
interpretations of a calculus, as for the positivists: rather, they are representations
that are particularly useful to human cognizers and their associated technologies.
Here Giere cites Eleanor Rosch and her associates on prototypes. 20 The
traditional, axiomatic view of theories assumes the classical analysis of concepts in
terms of logically necessary and sufficient conditions, whereas prototype theory
takes a more rhetorical line (my term again), viewing a concept (roughly) as a
18 Note that the basic models themselves are decreasingly considered applications of the theory. A
hierarchy of models comes into view. Cf. Suppes 1962, 1974.
19 For my own account of Kuhn's theory of cognition, see Nickles 1998,2002.
20 See, e.g., Rosch 1973 and Rosch and Mervis 1975.
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 15
similarity mapping from one or a few prototypical items to more peripheral items
classified together on grounds of family resemblance. So for Giere the structure of
theories reflects the structure of human cognition rather than simply and directly
expressing the representational content of reality. A theory and its models function
as tools, and a tool such as a crowbar is effective only insofar as it fits the world at
one "end" and human cognition at the other "end". (The crowbar is my own
analogy.)
Giere is sympathetic to the Kuhnian idea that human cognition is based on
pattern matching rather than on logical rules of either scientific method or an innate
language of the brain. Giere (1994, 278n2) locates his paper as "part of a larger
project to develop a more substantial cognitive foundation for a model-theoretic
approach to understanding scientific theories". His model-theoretic view of theories
is closer to scientific practice both as exhibited by detailed historical and
sociological studies of actual science and as revealed in accounts of human learning
and cognition from the cognitive sciences. In principle, he allows, this latter sort of
approach could yield normative claims that conflict with actual scientific practice.
The radical nature of Giere's departure from the traditional conception of
theories is evident in his recent comment on the cognitive development of children.
The plausibility of attributing theories to young children is strained by assuming that
theories are sets of statements. Children's linguistic abilities are limited. It is difficult to
explain the internal coherence of children's theories in terms [of] the logical coherence
of sets of statements. Nor is it very plausible to understand prediction and explanation
by children as involving implicit deductions. These difficulties vanish on a model-based
account of theories. The models associated with scientific theories have many of the
features of the "mental models" developmental psychologists wish to attribute to
children. The structure of a scientific model is independent of any particular linguistic
formulation and, indeed, may be independent of any linguistic formulation
whatsoever-as in the case of theoretical models characterized by reference to scale
models or diagrams. Moreover, the application of a model to any particular real
situation has the character of applying a predicate to that situation, something even very
young children obviously can do, as in "doggie."
Working within the perspective of a model-based account of scientific theories,
therefore, I have no difficulties whatsoever with attributing theories to children. [Giere
1996,538f]
Notice that Giere here abandons the traditional equation of cognition or thinking
or reasoning with logical inference. Indeed, thinking things through does not even
require language. It may involve running one or more mental models of a situation
in a kind of mental simulation. People working in inconsistency-tolerant logics may
attempt to compromise by urging that thinking can involve logical inference of
nonclassical varieties (Meheus 1999).
What does consistency amount to on Giere's model-theoretic view of science?
He perhaps assumes that all legitimate models, including mental models employed
by non-linguistic creatures, can be articulated as predicates in some language. 21 For
21 This is a controversial assumption, basically that all thought can be expressed in language, that
anything that can be thought can be said, a version of linguistic nominalism. On a still more
thoroughly pragmatic view, thought is not intrinsically mental but rather is expressed in practices of
all kinds.
16 THOMAS NICKLES
22 The structuralists impose "constraints" that require a certain coherence in the application of models.
23 On the other hand, Cartwright (1997) speaks of models as "blueprints for laws"!
24 A more complex case is Stephen Toulmin, a nonpositivist who hewed a Wittgensteinean, "family
resemblance" line that anticipates some of the moves described above. He applied Wittgenstein's
dictum-"Don't ask for [context-free] meaning, ask for use"-to scientific statements. Toulmin
(1953, chap. 3) treated law claims as "inference tickets", a view that is not unproblematic but that
anticipates later treatments of lawlike claims as nonrepresentational, practical tools. Accordingly, he
denied that laws are universal in scope or strictly true or false in general. Of a law statement we
should ask not "Is it true or false?" but "Where does it apply?" In his later Human Understanding
(1972), Toulmin historicized and biologized his conception and spoke of historical genealogies of
exemplary problems and solutions. He attacked what he saw as the over-systematized conception of
science characteristic of the positivists but also of Kuhn; and he proposed instead a looser,
evolutionary conception of scientific change.
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 17
grounds. However, this fear of the Ptolemaic seems unwarranted, for several
reasons. First, given the deflation of theories to a loose collection of models with
little or no underlying formal linguistic and deductive structure, it is no longer
perfectly clear where a charge of inconsistency is to be laid. Is it an inconsistency in
the representational content of different models (which, after all, are ideal models)
or an inconsistency in the practice of applying them to real-world systems?
Syntactical and even semantic consistency no longer have their old purchase, and we
are thrown back upon pragmatic inconsistency (whatever, exactly, that is).
Second, on Giere's account (as with other rhetorical accounts), there are degrees
of applicability, corresponding to degrees of approximation. What, then, does it
mean to say that all applications of a given theory must be mutually consistent? In
practice scientists use whatever model can readily be applied to achieve the desired
degree of approximation. Human cognitive manageability as well as the nature of
the problem itself typically figures in the choice of model. Ask one question about a
metal rod and it is treated as rigid and continuous. Ask another question and it is
treated as a lattice of discrete atoms; and so on. 25
Third, the relation of models to reality is further complicated by the fact that
there often exists a hierarchy of models. The more theoretical models are not applied
directly to reality in the form of raw data but to phenomenal models of the data that
are in turn based on models of the experiment. Suppes (1962, 1974) already showed
that scientists construct hierarchies of models of data, that is, relational structures
that model the experimental data in ways that counterfactually abstract away from
certain local contingencies. Like the construction of theoretical models, however,
there is scope for creativity here, so we must expect that variant models will often be
possible. Just as there is a problem of finding applications of theories to real-world
phenomena, so there is a problem of constructing the data into phenomenal patterns
that will be relevant to extant theoretical models. More recently Bogen and
Woodward (1988) have made a similar point in their distinction of phenomena from
data.
Finally, on a more practice-centered conception of science, models are not top-
down derivations from theories conceived as comprehensive logical systems but
(often) bottom-up, purpose-relative attempts to model concrete phenomena.
Theories and high theoretical models may playa minor role in this process.
While Giere's account is non-Copernican in the ways that I have indicated, of all
the philosophers cited in this section it is Nancy Cartwright who comes closest to a
Ptolemaic conception of theories, laws, models, and their applications. In How the
Laws of Physics Lie (1983, Ilff et passim), Cartwright contended that, contrary to
the traditional view, even in physics (or especially in physics),
it is usual to give alternative theoretical treatments of the same phenomenon. We
construct different models for different purposes, with different equations to describe
them. Which is the right model, which the 'true' set of equations? The question is a
mistake. One model brings out some aspects of the phenomenon; a different model
brings out others. Some equations give a rougher estimate for a quantity of interest, but
are easier to solve. No single model serves all purposes best. ... [po 11]
25 There is a large literature on approximation, idealization, abstraction, simplification, etc. For a start, see
Shapere 1984, chaps. 2, 17, and Laymon 1985.
18 THOMAS NICKLES
[T]here are no rigorous solutions for real life problems. Approximations and
adjustments are required whenever theory treats reality .... [po 13]
I propose instead a 'simulacrum' account. ... [in which] to explain a phenomenon is to
construct a model which fits the phenomenon into a theory. The fundamental laws of
the theory are true of the objects in the model, and they are used to derive a specific
account of how these objects behave. But the objects of the model have only 'the form
or appearance of things' and, in a very strong sense, not their 'substance or proper
qualities'.
The covering-law account [of explanation] supposes that there is, in principle, one
'right' explanation for each phenomenon. The simulacrum account denies this. The
success of an explanatory model depends on how well the derived laws approximate the
phenomenological laws and the specific causal principles which are true of the objects
modelled. There are always more phenomenological laws to be had, and they can be
approximated in better and in different ways. There is no single explanation which is the
right one, even in the limit, or relative to the information at hand. Theoretical
explanation is, by its very nature, redundant. [po 17]
However, Cartwright did say that causal explanation is different. "We do not tell
first one causal story and then another" (p. 11). She remained a causal realist.
Cartwright's overall position was expressed by the provocative title of her book: if
we understand laws and theories in the traditional way, then we must say that the
laws and theories of physics lie. They purport to provide a true, unified picture of
nature, but in fact they do not. They are false, and they adequately handle, to a
useful degree of approximation, only a few, scattered, natural systems. 26
In more recent work (e.g., 1995, 1999), Cartwright rejects the deductive
covering-law model more firmly and relegates theories to an even more subsidiary
role in scientific practice. Theories and laws do not lie, because they do not carry
representational content at all. They are better seen as tools in the toolbox of science,
along with many other kinds of tools that help us construct models of interesting
phenomena. Neither Newton's laws, the SchrOdinger equation, nor some more
abstract Hamiltonian, known or unknown, correctly describes nature as a whole, or
describes any real system at all. Cartwright characterizes her position as
instrumentalist, but it is a broad instrumentalism with regard to entire fields of
inquiry and not limited to an instrumentalist interpretation of theories.
The traditional view that there must be a single, universal, representational
theory of nature, since nature itself is (supposedly) complete, unified, and consistent,
embodies an unjustified metaphysical presupposition about the unity of nature that
Cartwright challenges explicitly. Her attack on the unity of nature can be construed
as an attack on both traditional presuppositions, (1) that nature is unified and (2) that
nature is consistent and rational, in the sense that there must exist any master theory
(let alone a unique master theory) that is complete, consistent, and rationally
intelligible. Here is the opening paragraph of Cartwright's latest book, The Dappled
World (1999), which focuses attention on both physics and economics:
This book supposes that, as appearances suggest, we live in a dappled world, a world
rich in different things, with different natures, behaving in different ways. The laws that
describe this world are a patchwork, not a pyramid. They do not take after the simple,
elegant and abstract structure of a system of axioms and theorems. Rather they look
To sum up: the change from the traditional conception of theories to the various
semantic and Cartwrightian conceptions that devalue high theory in favor of lower
models runs deep. On the older view, theories are absolute, unqualified depictions of
reality. They are products of scientific practice, of course, but need not reflect that
practice or bear any special relationship to human cognition. Models are true,
intended interpretations of the theory calculus and, so, participate in the theory's
absolute generality. By contrast, the structuralists and especially Giere and
Cartwright, construe models as purpose-relative rather than absolute. No one model
captures everything we might want to know about the phenomena or systems in
question. To be sure, models may serve a representational function (albeit one more
local or phenomena-centered than that of old-style theories), but they also reflect
human interests and human cognitive abilities. A model gives us one angle on the
world, one "take" or point of view, one way of grabbing onto the world, but not the
only one. And it is models, by and large, rather than abstract theories, that embody
the computational resources.
It seems fair to describe the shift from theories to models as a shift from an
absolutist, "God's eye" account of science (or its ultimate goal) to a more pragmatic,
perspectival, human-centered account, and, correspondingly, a shift from a high
logical account to a more informal and rhetorical account.
It is pretty clear what consistency and inconsistency mean within a comprehen-
sive world picture that is logically regimented. But what do they amount to on more
perspectival and relativistic conceptions of human inquiry, where syntax-based
accounts give way to accounts based on semantics and pragmatics, that is, on
meaningful abstract structures and human intentions and practices of applying them?
On the one hand, the opportunity for inconsistency (through clashes of perspective)
increases greatly. On the other hand, it is no longer clear where and when such
clashes are cognitively debilitating, for the very point of a perspectival view is to
allow multiple perspectives.
centuries, however, we have learned that relaxing these requirements does not
necessarily bring epistemic disaster anymore than it brings political disaster.
Although consistency remains a strong desideratum, and justifiably so, consistency
increasingly becomes a regulative ideal for the "final" products of research rather
than something to be imposed rigidly from the start on the process of research. For,
again, a fallible, nonmonotonic process can hardly be temporally consistent.
Even inconsistency in the final products can often be localized and thereby
tamed. Inconsistency between theories in neighboring fields, such as that between
late 19th-century physics and geology, is undesirable yet tolerable in the sense that
work in the two fields does not grind to a halt. This is fortunate since inconsistency
is a symmetrical relation that does not indicate which (if either) of the two "sides" is
"right", or on a better track.
The same point can be made, somewhat more cautiously, about inconsistency
within individual fields. One thinks of relativity and quantum theory, for instance.
Research continued fruitfully in both fields for decades, despite recognition of their
mutual incompatibility. However, this incompatibility has now become an obstacle
to progress in constructing a unified theory. Indeed, the demand for consistency is
now a powerful constraint on theorizing in this field. Even the pragmatists among us
can appreciate the concern of Weinberg and many others with this inconsistency,
since it blocks the road to potentially fertile research. It suggests that there is
something very basic about the world that we do not understand. No, worse. It is not
simply that there is a basic, brute fact for which we have no explanation. In this case
the inconsistency strongly suggests that we deeply misunderstand our world.
When we look closely at scientific practice, including the construction of models
and theories, we not infrequently find toleration of inconsistency even within
individual theories and research practices, at least in their nascent phases, or as long
as they continue to promise interesting results.
The point here is that heuristic fertility trumps consistency. Researchers rarely
reject a promising approach on the ground that it is apparently inconsistent.
Conversely, inconsistency does not necessarily kill all promise of problem-solving
success. Not all inconsistencies are equal. Some are profound while others are mere
nuisances, rough spots to tiptoe around. In the latter case, achieving consistency can
be a verbal nicety that researchers relegate to the level of "mop up work" (more
guarded phrasing of claims, etc.). In sum, actual practice often reverses the
traditional view of philosophers and logicians, that consistency trumps all other
constraints, and certainly any "merely heuristic" considerations.
Another, more familiar point is that, still within a given field, conflicts between
the theories or models and the phenomena are frequent. We have already rehearsed
Kuhn's (1970a) rejection of Popper's position on falsification and his claim that
normal science can tolerate anomalies up to a point. 27 (To be sure, internal
consistency remains a desideratum, since anomalies become one focus of normal
27 Popper himself, and his students, claimed that inconsistency lies at the core of all problems. (See
Hattiangadi 1978). They had chiefly in mind clashes of theory with observation and clashes of one
theory with another. In tum, these clashes often reflected metaphysical antinomies, e.g., between
continuous and discrete representations of reality.
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 21
research, that aims to eliminate them.) Most science studies experts, apparently
including Kuhn himself, would in tum reject the claim that "normal" scientific
communities are tightly bound by consensus (and hence consistency) of belief
(Fuller 1989, Rouse 1996). Finally, there are many examples of scientists who
opportunistically employ incompatible ideas within a single theory, indeed, within a
single paper. In his first paper on Brownian motion, for example, Einstein (1905b)
simultaneously employed two incompatible models of a fluid in order to derive an
important result (Brush 1968, 18; Nickles 1980).
Where does this leave the question of when and where it is permissible to violate
the consistency constraint? The question reduces to that of when and where heuristic
appraisal, including matters of cognitive economy, overrule consistency demands.
Unfortunately, this area of methodology remains very underdeveloped.
Probably the most extensive discussion of inconsistency in research is that of
Nicholas Rescher and Robert Brandom. In The Logic of Inconsistency (1979), they
take the radical step of exploring and defending the idea that nature itself might be
inconsistent in the sense that any adequate description requires a logic that includes
nonstandard models. All of the authors featured in the previous section would
challenge the tacit assumption, underlying much scientific writing from Copernicus
to Weinberg, that there is a single, correct, "God's-eye" view of the universe that
can in principle be captured by a single, unified, master theory; but Rescher and
Brandom go further.
Whatever the warrant for their radical thesis, in my opinion Rescher and
Brandom are correct to say, only slightly less radically, that, generally speaking, we
should regard consistency as one important constraint among others rather than as
a non-negotiable prerequisite for productive inquiry. This means that it may be
fruitful (and hence permissible or "rational") in the heat of research to violate the
consistency constraint, just as it may be fruitful to violate any other constraint. 28 In
short, good work may be inconsistent with the consistency constraint! Yet many
20th-century logicians and epistemologists still assign to inconsistency an infinite
negative utility value in the economy of research, which implies that inconsistency
never pays.
Permitting independence of thought and practice, and hence risking inconsis-
tency via disagreement, is a necessary condition for successful inquiry in any area
where absolute proof is unattainable; for in those domains we have nothing more
reliable than methods and results that are robust or consilient. Each variety of
robustness amounts to a different way in which independent paths converge to the
same result (Wimsatt 1981). If two people agree simply because one defers to the
other as an authority, that agreement is epistemically uninteresting, even pernicious;
whereas, when independently thinking persons employing independent methods
arrive at the same conclusion, that's news. The same point holds if a theory survives
simply because any proposal counter to it is rejected out of hand rather than, say,
being the conclusion to which its predecessors converge in some limit.
28 Constraint violation itself mayor may not be a form of inconsistency, a charge that is somewhat
negotiable; see Nickles 1980 and below.
22 THOMAS NICKLES
I conclude that (1) certain kinds of inconsistency must be tolerated during the
research process right up to the final stages of refining the product; and (2) it is
usually better for consistency of results to emerge as an achievement of the
investigative process rather than to be imposed rigidly on the process itself. When
won under risk of inconsistency, consistency of results becomes a genuine
achievement and is itself a mark of robustness. (A similar claim can be defended for
stronger forms of coherence and consilience than mere consistency.) Accordingly,
methodology should treat this sort of consistency as an aim, as a desired feature of
the refined conclusions of research, rather than as a prior condition of research itself.
Overzealous consistency requirements are idle in any case, rather like the old
view that any knowledge claim that is less than certain is worthless. Foundationist
programs, from Cartesianism to some versions of operationism and positivism, have
imposed the demand that inquiry proceed carefully enough to identify and avoid all
errors in advance. Historical, logical, and epistemological work over the past several
decades has taught us that this demand is impossible to meet even for the gross
"error" of inconsistency. Perfect consistency can rarely be guaranteed, for well
known logical and computational reasons. The consistency of interesting logical and
mathematical systems is usually unprovable except relative to some other system.
And a "Cartesian" monitoring of the consistency of our belief sets is impossible.
Cherniak (1986, 93) shows that even the entire universe operating as a computer
could not monitor, by truth-table methods, the consistency of a belief set containing
only 138 logically independent propositions. To put this point in historical
perspective: meeting the consistency requirement is just as impossible as meeting
the Baconian and Cartesian methodological requirements that we begin inquiry by
emptying the mind of all "prejudice", that is, of everything we think we know.
To take a more recent example, a requirement of total consistency is just as
impossible to meet as the requirement of total evidence in positivist accounts of
inductive inference, and for a similar reason. The total evidence requirement was an
attempt to avoid inductive inconsistencies, given that inductive inference is
nonmonotonic (Hempel 1960).
Not only is the consistency requirement not rigorously enforceable in practice
but also scientists sometimes find it fruitful to violate it deliberately, by "accepting"
(for the purpose of serious research) claims that are known to be inconsistent with
"established" knowledge. In another of the famous 1905 papers, the "quantum
theory" paper, Einstein (1905a), for definite reasons, employed Wien's distribution
law for blackbody radiation even though he knew Wien's law was false and in fact
inconsistent with Planck's formula, which had been empirically confirmed (Nickles
1980).
The seriousness of inconsistency charges always depends on what else is at
stake. Consistency can, of course, be a major, active constraint on research, focusing
inquiry in certain directions rather than others. (As noted above, this is a basic
argument that realists since Copernicus and Kepler have employed against
nonrealists and that Popper's "critical approach" exploits.) Still, I defend the
Rescher-Brandom claim that consistency is often one constraint among others and
not necessarily the most important one at all stages of research.
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 23
But does not inconsistency guarantee that your claim is false, and is this not
important in itself? Yes and no. Yes the claim is false; but no, this is not decisive,
for false models or hypotheses can lead to better ones. The history of science affords
a host of examples, our recent Einstein case being one. Wimsatt (1987) discusses
others.
What is at stake is relative to context and to purpose. Charges of inconsistency
can be logically and rhetorically negotiable, a matter of tradeoffs with other
desiderata or commitments within a system of epistemic and social accounting.
Because this is so, consistency can often be considered a soft constraint rather than a
hard constraint. Here are more examples.
Few if any late medieval and renaissance mathematicians would have character-
ized Ptolemaic astronomy as inconsistent. It is only by contrast with the
epistemologically optimistic, strongly realist-representational world picture
advanced by Copernicus and Kepler that Ptolemaic practice looks so disjoint. 29 To
be sure, in historical retrospect the imposition of physical realism and the
corresponding consistency constraints did help to advance inquiry in that context.
These moves turned out to be very fruitful. However, if we give any credence to the
deflationary positions of §4, this is not a universal lesson. From the point of view of
contemporary practice in the physical sciences (not to mention the biological and
other sciences), as interpreted by Kuhn, Giere, and Cartwright, Ptolemaic practice is
partially rehabilitated. 30 And all three of these writers claim to be realists of one sort
or another! To take a more recent example, most American physicists fruitfully
employed the new quantum theory without getting embroiled in debates about its
realistic interpretation, completeness, or consistency (Cartwright 1987, Krips 1996).
Furthermore, consistency and full-blooded realism are sometimes at odds. This is
a familiar point in the interpretation of quantum theory, but let me revert to a
previous example. Kepler himself played off consistency against realism when the
going got tough. What was his response to the problem, explicitly raised by
Osiander and Ursus, that a system of the world appropriately based on epicycles and
another system based on eccentrics, can be mathematically equivalent (and hence
would save all the phenomena no matter how much observational precision was
improved) yet are physically inconsistent? At this point Kepler, in order to save
consistency, retreated from his full-tilt realism to saying that these were just two
different mathematically conventional ways of expressing the same physical reality,
in other words that we need not interpret these epicycles and eccentrics as real. 31
Inconsistency problems can often be tamed by making sometimes minor
adjustments elsewhere. Quine's web-like system of centralities and priorities comes
to mind here. A common response to inconsistency is to deny that what appears to
29 Actually, something of this claim could already be made on the basis of Copernicus's streamlining of
computational practice. Whether Copernican calculation was generally easier than Ptolemaic, I do not
know. Commentators make claims both ways, possibly pertaining to different problems.
30 Again, I don't want to overstate this point. Cartwright, for example, is a realist when it comes to causal
mechanisms.
31 This is part of a more complex story. See Jardine (1984, 219), who, inter alia, places more emphasis on
the debate between epistemic optimists and pessimists and less on the debate over realism.
24 THOMAS NICKLES
be P and not-P are really contradictories, or even contraries. "Whenever you meet a
contradiction, you must make a distinction", writes William James (1907, chap. 2).
Here is a range of examples of evading an inconsistency charge by drawing a
distinction, that is, by relativizing the claims to the different domains on either side
of the distinction. "It is both raining and not raining." "Yes, but at different times
and places" (relativity to time and place). A politician has declared himself both for
and against a measure. "Yes, but at different times and in response to different
information" (relativity to premise set and, perhaps, goal).32 "You say that rock
group is both hot and cool." "Yeah, man, but those words mean the same thing here"
(relativity of language to context). "You are both pleased and displeased with my
work." "Yes, but in respect to different pieces (or aspects) of the work." When we
chase a squirrel around a tree but never catch up to it, we are running around the
squirrel, and yet we are not (James 1907, chap. 2). "Yes, from two different
perspectives, or in two different senses of the word 'around'." "Your theory cannot
consistently explain this data in the same way that it explains that data." "Yes but I
only claim that my theory explains the phenomena, not the raw data" (Bogen and
Woodward 1988). "You treat this metal rod as an atomic lattice, yet you also treat it
as continuous." "Yes, depending on the particular application" (relativity to problem
context). "You accept Einsteinean relativity and yet you deny that classical
mechanics is false." "Yes, classical mechanics remains true within the domain of
low velocities, etc., which concerned Newton and his successors." "Light is
wavelike, yet it is particulate." "Yes, but relative to different experimental setups"
(Niels Bohr).
Such distinctions may be rather fine. In any case one then must face the
responsibility of convincing one's audience that the distinction is legitimate and not
purely ad hoc logic-chopping, for "any difference to be a difference must make a
difference". But a difference to whom, where, when, why? To advance a distinction
is to make a content-laden claim about the world in relation to our practices and
goals. Whether or not the relevant audience judges the distinction a good one is not
only a question of what we know about the world but also a matter of heuristic
appraisal and economics-of whether or not the relevant community thinks it is
worth making.
Every major distinction made in philosophy and the sciences can function as a
consistency-preserving device. It would be interesting to examine the history of
philosophy and the history of scientific inquiry from this angle. Hegel, for instance,
challenged many traditional philosophical dualisms and dichotomies. He was
accordingly more tolerant of contradiction. In fact, he saw it as a kind of creative
force and the clue to a new sort of logic of change. Hegel's was a conflict theory of
inquiry quite interestingly opposed to the monotonic, linear-foundational accounts
that had preceded him. Although Hegel did blur many dualisms, he managed
inconsistency by drawing new distinctions of his own.
32 Even your doing X and then later doing Y in exactly the same circumstance need not be inconsistent in
a morally or epistemically pejorative sense. Perhaps you chose the parfait for desert yesterday and the
cake today, simply for variety. Perhaps there was a change of desire, but even that does not really
matter in this case.
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 25
33 Wimsatt has treated these themes in a series of articles over the years. For recent examples, see
Wimsatt 1981, 1987, 1997, forthcoming.
26 THOMAS NICKLES
The moral, political, and intellectual history of the West has seen a gradual
weakening of traditional assumptions about the requirements for a viable society as
well as for useful knowledge. No longer do we accept the dictum of the Sun King,
Louis XIV, that a stable society requires un rai, une lai, une fai. No longer do we
fear that the social order will collapse if the working class, or blacks, or women are
given the vote.
In theory of inquiry, as noted above, we have also been loosening the bonds of
consistency since the time of Descartes and Hobbes, both of whom abhorred
diversity and relativity as the roots of skepticism and factionalism. No longer do we
impose Platonic and Cartesian dichotomies between invaluable, absolute,
unassailable episteme on the one hand and virtually worthless doxa on the other. No
longer do we draw sharp, Platonic boundaries between theory and practice and
between method and content. Another way to put this point is that methodology has
moved ever further away from the omniscience or "God model" of knowledge and
inquiry. Theory of inquiry has become increasingly secularized and increasingly
fallibilistic. Some main stages in this development are these.
34 Ironically, however, an omniscient God need not be at all concerned with cognitive economy. Indeed,
an omniscient being cannot genuinely inquire (upon pain of the first hom of the Meno paradox). The
way out of this difficulty is to imagine that God created the world so that we finite humans can figure
it out. The universe is a big intelligence test.
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 27
35 Antecedents can be found in Bacon's refrain, "the art of discovery may advance as discoveries
advance", and in Peirce's 1877 "each chief step in science has been a lesson in logic". Shapere (1984)
strongly defends the "learning to learn" model. My survey is, of course, oversimplified. E.g., Lakatos
(1970) attempted to incorporate self-correction methods within a fixed methodology. See also his
study of mathematical reasoning (Lakatos 1976).
28 THOMAS NICKLES
The more kinds of incompatibility represented in the initial population, the more
severe the competition, and the higher the probability of arriving at a good solution
sooner rather than later. Note, incidentally, that, on this approach, again, not all
inconsistencies are equal. By contrast, in standard symbolic logic once you have one
inconsistency you have them all, and it makes little sense to speak of diversity.
If evolutionary epistemologists such as Donald Campbell (1974), Richard
Dawkins (1986), Dan Dennett (1995), and Gary Cziko (1995) are correct, this point
applies not only to research communities but also to each individual problem solver,
at least at the subconscious level, where our brains are engaged in a kind of
evolutionary exploration of the problem space. For their claim is that a selectionist
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 29
account of learning and inquiry is the only defensible account of inquiry that we
possess. 36
Other burgeoning areas of artificial intelligence that are interesting from the
standpoint of consistency are case-based reasoning (CBR) and model-based
reasoning (MBR), both of which were anticipated by Kuhn on exemplars. In the
Postscript to the second edition of Structure, he contended, in effect, that scientific
inquiry is case based (or perhaps model based) rather than rule based, at least at the
conscious level. His claim, that expertise on sets of exemplars predicts future
scientific behavior better than sets of rules, is increasingly supported by work in
cognitive science and science studies. Just as Giere interprets Kuhn's most canonical
exemplars as models, CBR theorists interpret them as cases in a case library. Similar
questions arise for CBR as we raised for MBR above, concerning the definition,
practical identification, and importance of inconsistency in such systems.
Computerized search for cases in large databases has been used in law for some
time, and more sophisticated CBR systems are now being developed. One
interesting feature in the legal sphere is that legal reasoning in the United States and
several other countries is typically adversarial, meaning that there are two opposing
sides, each trying to outdo the other in finding the closet legal precedents (case
matches) for the case they are arguing. Since a good attorney will try to determine
not only her own best case but also the best case that the opposition can muster, the
most useful retrieval system will return the closest matches with opposite verdicts.
There is no real inconsistency here, but effective advocacy does require people and
systems that can handle incompatible points of view. Scientific reasoning is not
explicitly and automatically adversarial in this manner, although it can be
adversarial. In any event, anyone writing a scientific paper should, in the process,
play the role of the skeptic in trying to challenge the claimed results. 37
To be sure, computer implementations of CBR to date must use "logical"
computer languages. However, even here CBR frequently has advantages over
programs that are fully rule-based problem solvers (Nickles 1998). Case-based
systems often "scale" better than rule-based systems. As more rules are added, rule-
based systems become slower and less efficient rather than faster, at a more-or-less
exponential rate. In other words, their performance degrades (in this respect) rapidly
as they "learn" more. A main reason for this is that rule-based systems are less
monotonic than are case-based systems. The latter tend to be more distributed and
less strongly connected, logically speaking; hence, conflict or inconsistency does not
immediately propagate throughout the system. Accordingly, it is often easier to add
new cases to the case library of a case-based system than it is to add new rules to the
rule set of a rule-based system. New cases can conflict with old, but the conflicts are
36 The aforementioned authors contrast selectionist accounts, based on blind variation plus selective
retention, with providential (God-design) and instructivist (passive induction) accounts ofieaming.
37 For an entry into case-based reasoning, see Kolodner 1993 and Leake 1996. Adversarial thinking is one
clue as to how scientists and others respond to the problem that everything is similar to everything
else in some respect or other. See Ashley 1990 and compare Andersen 2000 on Kuhn's use of contrast
sets.
30 THOMAS NICKLES
usually not as frequent or as severe as when rules are added to a rule-based system. 38
Again we meet the tradeoff between the computational power of high deductive
integration, where it is genuinely available, and susceptibility to contamination or
pollution through immediate propagation of error.
Insofar as we can trust many recent studies, scientific practice is closer to the
Ptolemaic paradigm and further from the Copernican paradigm than the received
philosophical accounts would have us believe. I have not attempted any final
resolution of the debate here, although I personally find myself being a good
Copernican only on Sundays. I end the paper with a series of (overlapping)
questions rather than conclusions.
What direction will (or should) the consistency discussion and the corresponding
develop of logics take ...
38 See Kolodner 1993 and Leake 1996. Advocates of rule-based reasoning will contest some of these
claims.
FROM COPERNICUS TO PTOLEMY: INCONSISTENCY AND METHOD 31
11. if we take seriously the more deflationary and pragmatic accounts of scientific
work?
REFERENCES
Andersen, H. (2000), Kuhn's Account of Family Resemblance: A Solution to the Problem of Wide-Open
Texture. Erkenntnis 52, 313-337.
Ashley, K. (1990), Modelling Legal Argument: Reasoning with Cases and Hypotheticals. Cambridge,
Mass.: MIT Press.
Balzer, W., C. U. Moulines, and J. Sneed (1987), An Architectonicfor Science. Dordrecht: Kluwer.
Balzer, W., and C. U. Moulines (1996), Structuralist Theory of Science: Focal Issues. New Results.
Berlin: de Gruyter,.
Bogen, J., and 1. Woodward (1988), Saving the Phenomena. Philosophical Review 97,303-352.
Brush, S. (1968), A History of Random Processes I: Brownian Movement from Brown to Perrin. Archive
for History of Exact Sciences 5, 1-36.
Campbell, D. T. (1974), Evolutionary Epistemology. In The Philosophy of Karl R. Popper, vol. I, P. A.
Schilpp (ed.), La Salle, Ill.: Open Court, 1974, pp. 412-63.
Cartwright, N. (1983), How the Laws ofPhysics Lie. Oxford: Oxford University Press.
Cartwright, N. (1987), Philosophical Problems of Quantum Theory: The Response of American
Physicists. In The Probabilistic Revolution, vol. 2, L. Kruger, G. Gigerenzer, and M. Morgan (eds.),
Cambridge, Mass.: MIT Press, 1987, pp. 417-37.
Cartwright, N. (1997), Models: The Blueprints for Laws. Philosophy of Science 64, 292-303.
Cartwright, N. (1999), The Dappled World: A Study of the Boundaries of Science. Cambridge: Cambridge
University Press.
Cartwright, N, T. Shomar, and M. Suarez (1995), The Tool Box of Science. In Herfel et al. (1995),
pp. 137-49.
Cartwright, N., J. Cat, L. Fleck, and T. Uebel, eds. (1996), Otto Neurath: Philosophy Between Science
and Politics. Cambridge: Cambridge University Press.
Chamberlain, T. C. (1897), Studies for Students. Journal of Geology 5, 837-848. Reprinted under the
title, The Method of Multiple Working Hypotheses. In Philosophy ofGeohistory. 1785-1970, C.
Albritton (ed.), Stroudsburg, Penn.: Dowden, Hutchinson and Ross, 1975, pp. 125-131.
Chemiak, C. (1986), Minimal Rationality. Cambridge, Mass.: MIT Press.
Cziko, G. (1995), Without Miracles. Cambridge, Mass.: MIT Press.
Dawkins, R. (1986), The Blind Watchmaker. New York: Norton.
Dennett, D. C. (1995), Darwin's Dangerous Idea. New York: Simon & Schuster.
Dewey, J. (1908), Does Reality Possess Practical Character? In Essays. Philosophical and Psychological.
in Honor of William James. New York: Longmans, Green.
Duhem, P. (1969), To Save the Phenomena. Chicago: University of Chicago Press. Originally published
in 1908.
Dupre, J. (1993), The Disorder of Things. Cambridge: Harvard University Press.
Einstein, A. (1905a), On a Heuristic Viewpoint about the Creation and Conversion of Light. Translation
reprinted in The Old Quantum Theory, D. ter Haar (ed.), Oxford: Pergamon Press, 1967, pp. 91-107.
Einstein, A. (1905b), Investigations on the Theory of the Brownian Movement. Translation reprinted in
Investigations on the Theory of the Brownian Movement. New York: Dover, 1956.
32 THOMAS NICKLES
Magnani, L., N. Nersessian, and P. Thagard, eds. (1999), Model-Based Reasoning in Scientific Discovery.
New York: Kluwer.
Meheus, J. (1999), Model-Based Reasoning in Creative Processes. In Magnani et al. (1999), pp. 199-217.
Moulines, C. U. (1996), Structuralism, the Basic Ideas. Chap. 1 of Balzer and Moulines (1996).
Nickles, T. (1980), Can Scientific Constraints Be Violated Rationality? In Scientific Discovery. Logic.
and Rationality, T. Nickles (ed.), Dordrecht: Reidel, 1980, pp. 285-315.
Nickles, T. (1987), Lakatosian Heuristics and Epistemic Support. British Journalfor the Philosophy of
Science 38,181-205.
Nickles, T. (1988), Questioning and Problems in Philosophy of Science: Problem-Solving Versus
Directly Truth-Seeking Epistemologies. In Questions and Questioning, M. Meyer (ed.), Berlin:
Walter De Gruyter, 1988, pp. 38-52.
Nickles, T. (1998), Kuhn, Historical Philosophy of Science, and Case-Based Reasoning.
Configurations 6, 51-85 (special issue on Thomas Kuhn).
Nickles, T. (2002), Normal Science: From Logic to Case-Based and Model-Based Reasoning. In Thomas
Kuhn, T. Nickles (ed.), Cambridge: Cambridge University Press.
Peirce, C. S. (1877), The Fixation of Belief. Reprinted in Collected Papers of Charles Sanders Peirce,
vol. 5, C. Hartshorne and P. Weiss (eds.), Cambridge, Mass.: Harvard University Press, pp. 358-387.
Perez-Ramos, A. (1988), Francis Bacon's Idea of Science and the Maker's Knowledge Tradition. Oxford:
Clarendon Press.
Quine, W. V. O. (1951), Two Dogmas of Empiricism. Reprinted in From a Logical Point of View.
Cambridge, Mass.: Harvard University Press, pp. 20-46.
Quine, W. V. O. (1969), Natural Kinds. In Ontological Relativity and Other Essays. New York:
Columbia University Press, pp. 144-138.
Rescher, N., and R. Brandom (1979), The Logic of Inconsistency. Oxford: Blackwell.
Rosch, E. (1973), Natural Categories. Cognitive Psychology 4, 328-350.
Rosch, E., and C. B. Mervis (1975), Family Resemblances Studies in the Internal Structure of Categories.
Cognitive Psychology 7, 573-605.
Rouse, J. (1987), Knowledge and Power: Toward a Political Philosophy of Science. Ithaca: Cornell
University Press.
Rouse, J. (1996), Engaging Science: How to Understand Its Practices Philosophically. Ithaca: Cornell
University Press.
Schurz, G. (1995), Theories and their Applications-A Case of Nonmonotonic Reasoning. In Herfel et al.
(1995), pp. 269-94.
Shapere, D. (1984), Reason and the Search for Knowledge. Dordrecht: Reidel.
Stegmliller, W. (1976), The Structure and Dynamics of Theories. New York: Springer-Verlag.
Translation of the 1973 German edition.
Suarez, M. (1999), Theories, Models, and Representations. In Magnani et al. (1999), pp. 75-83.
Suppe, F. (1974), The Structure of Scientific Theories. Urbana: University of Illinois Press.
Suppe, F. (1989), The Semantic Conception of Theories and SCientific Realism. Urbana: University of
Illinois Press.
Suppes, P. (1962), Models of Data. In Logic. Methodology and the Philosophy of Science. Stanford:
Stanford University Press, pp. 252-261.
Suppes, P. (1974), The Structure of Theories and the Analysis of Data. In Suppe (1974), pp. 266-307.
Toulmin, S. (1953), The Philosophy of Science: An Introduction. London: Hutchinson.
Toulmin, S. (1972), Human Understanding. Princeton: Princeton University Press.
Van Fraassen, B. (1987), The Semantic Approach to Scientific Theories. in The Process of Science, N.
Nersessian (ed.), Dordrecht: Martinus Nijhoff.
Weinberg, S. (1994), Dreams of a Final Theory. New York: Random House.
Westman, R., (1975), (ed.), The Copernican Achievement. Berkeley: University of California Press.
Wimsatt, W. (1981), Robustness, Reliability and Overdetermination. In Scientific Inquiry and the Social
Sciences, R. Brewer and B. Collins (eds.), San Francisco: Jossey-Bass.
Wimsatt, W. (1987), False Models as Means to Truer Theories. In Neural Models in Biology, M. Nitecki
and A. Hoffman (eds.), Oxford: Oxford University Press, 1987, pp. 23-55.
Wimsatt, W. (1997), Aggregativity: Reductive Heuristics for Finding Emergence. Philosophy ofScience
64,372-384.
Wimsatt, W. (Forthcoming), Emergence as Non-Aggregativity and the Biases of Reductionisms. In
Natural Contradictions, J. Haila and P. Taylor (eds.).
ARTHUR 1. MILLER
One of the most fascinating aspects of scientific creativity is how consistent theories
can sometimes emerge from inconsistent premises or inappropriate data. Using cases
from the history of science I will explore this process and try to draw some general
lessons concerning how science is done and, more particularly, about the concept of
scientific progress.
I will examine three important sources of inconsistencies:
We will see how recognising and then eliminating inconsistencies that arise from
sources of the sorts in statements 1-3 can lead to a better understanding of scientific
creativity.
This can also go under the heading "Unwanted Precision". Sometimes less accurate
data are more informative. Examples from Galileo's research are particularly
relevant here.
A thought experiment on which Galileo spent a great deal of time in his 1632
Dialogue on the Two Chief World Systems involves someone who, standing on the
mast of a moving ship, drops a stone (Galileo 163211967, 148-150). Galileo
pondered where it will land? Aristotelians argue that the ship moves out from under
the stone and so it does not fall directly under the person who dropped it. This is the
result expected from our intuition which, in an untutored state, is essentially
Aristotelian. Galileo, on the other hand, argues that the stone will land directly
beneath the person who dropped it.
Imagine, however, that Galileo had extremely precise measuring instruments, or
that he lived on a planet that spun on its axis orders of magnitude faster than ours.
He would have found that the stone does not fall directly under the person who
35
J. Meheus (ed.), Inconsistency in Science, 35--41.
© 2002 Kluwer Academic Publishers.
36 ARTHUR I. MILLER
dropped it. Its deviant course is due to a force whose origin is in the Earth's rotation
about its axis: the Coriolis force. The effects of this force depend on where you are
on the rotating Earth and how fast you are moving. The Coriolis force is of key
importance in the motion of large air masses such as cyclones, tornadoes, and trade
winds. But for falling bodies on the Earth, its effects are small. In Galileo's
experiment the deviation from vertical fall due to the Coriolis force for an object
dropped from a mast of height 30 feet at a latitude of 45° is only .02 inches. Luckily,
Galileo did not have highly sensitive measurement devices. Any measurements
revealing this deviation would only have confused matters because a first step in
understanding how objects fall is to formulate a theory of motion in which the
Coriolis force is not present. This is what Galileo did by restricting all considera-
tions to inertial reference systems and free fall through vacuum.
There is yet another force arising from the Earth's rotation called the centrifugal
force, which depends only on where you are on the rotating Earth. One of its effects
is to produce a variation in the acceleration of free fall depending on your latitude.
Any corrections are in the third decimal place.
Similarly, Galileo's experiments with pendula may well be flawed because he
pulled the pendulum bob far enough from the vertical so that the motion was not
simple harmonic. Better clocks for timing purposes would have been deleterious to
his goal of using pendula to convince himself that all bodies fall with the same
acceleration in vacuum regardless of their weight. This hypothesis is central to
Galileo's theory of motion.
We can say that Galileo reasoned toward his theory of motion with data that
were inexact and sometimes with fallacious logic. Besides his unbridled enthusiasm,
we can conclude only that most of these experiments served merely as checks on
already performed thought experiments and already accepted universal laws.
It is interesting to take note of the fact that a concept of key importance in the
theories of Einstein, Galileo and Newton, is the inertial reference system. This is a
platform that moves at a constant velocity in a straight line forever. We assume the
Earth is an inertial reference system. Yet the Earth hurtles around the sun at the rate
of one revolution per year at a mean speed of 67,000 miles/hour, while revolving on
its axis once every 24 hours at a mean rate of 1,040 miles/hour at the equator.
Despite these accelerated motions, engineers and scientists can still do the vast
majority of their calculations using Newton's laws as if the Earth were moving in a
straight line at a constant velocity. We have already noted corrections due to the
Coriolis and centrifugal forces. Any effects of the Earth's orbital acceleration on
laboratory experiments performed on the Earth depend on terms in Newton's
gravitational law containing ratios of the Earth's radius to that of the Earth's
distance from the sun. Detecting such effects requires accuracy to at least six
decimal places. All of this is fortuitous, because if it were not the case then it would
have been very difficult, indeed, for Galileo to have formulated a theory of motion
alternative to Aristotle's.
Since Newtonian science is based on inertial reference systems, we would expect
to see examples of them. Yet there are none in practice, only in principle, that is, in
the mind of the thought experimenter. We are faced with a situation in which the
very basis of Newtonian mechanics doesn't exist. Is this not inconsistent? We say
INCONSISTENT REASONING TOWARD CONSISTENT THEORIES 37
that Newton's theory is approximate, and don't blame the collapse of bridges on
there being no real inertial reference systems. Einstein's special theory of relativity
is also based on the inertial reference system in the sense that it deals only with
measurements made in such systems, which Einstein himself took to be its "logical
weakness" (Einstein 1923, 480). And so special relativity is also approximate.
Einstein repaired this situation in 1915 with his general theory of relativity in which
measurements can be made in accelerating reference systems.
This brings me to the ether-drift experiments of the latter part of the nineteenth
century. These were state of the art experiments performed by some of the greatest
high precision experimenters of the day. They sought to detect effects of the Earth's
motion through the ether on measurements of the velocity of light. The effects
sought were of the order of (v/c) = 10-4 and (V/C)2 = 10- 8 where, for the sake of
determining an upper limit, v is taken as the Earth's orbital velocity about the sun
(30 km/sec) and c is the velocity of light as measured in the free ether, that is,
vacuum, and is 3 x 108 m/sec. Yet are not the very premises of the (v/c) experiments
flawed? The reason is that this effect will be swamped by effects due to the Earth's
not being an inertial reference system, which can also affect second order results.
For example, in their famous 1887 ether-drift experiment, Michelson and Morley
took account only of the Earth's orbital motion about the sun. Their corrections,
however, were less than precise because they had to be folded into the effects due to
the motion of the solar system about which, in Michelson and Morley's words, "but
little is known with certainty" (Michelson and Morly 1887, 281). Moreover, had
these experiments or the subsequent more precise ones performed in the 1920's
succeeded, they would have been evidence neither for an ether nor against Einstein's
special theory of relativity. Rather, they would simply be indicators of the
"noninertiality" of the Earth as a moving platform.
Another example of unknowingly reasoning from restricted data is the Dutch
physicist H. A. Lorentz's successful explanation in 1896 of his colleague Pieter
Zeeman's data on the splitting of certain spectral lines of sodium in an externally
imposed magnetic field. Fortuitously, most of Zeeman's experiments occurred in
large enough magnetic fields so that, to use quantum mechanical terminology, the
spin and orbital angular momenta decouple. In this regime Lorentz's classical
electrodynamics sufficed for counting line splittings and approximately fitting line
spacings. Although in 1898 the so-called anomalous Zeeman effect began to be
noticed, in 1902, Lorentz and Zeeman were awarded the Nobel Prize in physics for
explaining Zeeman's high field data.
We tum next to a deeper level of inconsistencies.
What to do? The first step physicists took was to exclude light quanta from
atomic theories. This had been Niels Bohr's tack in 1913 and he continued it as long
as possible into the 1920's (see Miller 1986). Bohr's reason was the same as
Planck's-the "image of light quanta precludes explaining optical interference"
(Bohr 1921, 241-242). In the face of data favouring light quanta taken in 1923 by A.
H. Compton, Bohr offered a desperate attempt to exclude them by proposing in
1924, a version of his atomic theory that violated (inconsistent with) energy and
momentum conservation. Although it failed empirically, one of its fundamental
assumptions-virtual oscillators-turned out to be of great importance in
Heisenberg's discovery of the quantum or matrix mechanics in 1925 (see Miller
1994,4-8).
By 1927 the wave-particle duality moved from being inconsistent to paradoxical.
It seemed to be like relating apples and fishes. Consider a light quantum. Its energy
E and momentum p are related to its frequency v and wavelength A. through Planck's
constant h as
E = hv (1)
p = hlA. (2)
Eqs. (1) and (2) relate the light quantum's "particle" properties (E and p) with its
"wave" properties (v and A). Is this not inconsistent, that is, paradoxical. Bohr's
approach to paradox busting was unique: instead of reconciliation, or rejecting one
hom of the dilemma, he realised that both horns had to be embraced. As Bohr put it
in his complementarity principle of September 1927, taking serious account of
Planck's constant-which is the connector of the light quantum's "wave" and
"particle" modes [see Eqs. (1) and (2)]-means that we must extend our intuitivity
into a domain where, in fact, entities are "wave" and "particle" simultaneously
(Bohr 1928). In this regime the terms "wave" and "particle" do not have the same
connotation as they do in the world of sense perceptions. This is why I placed them
in quotes. In Bohr's own words (Bohr 1928, 590; Miller 1996,65-68):
Indeed we find ourselves here on the very path taken by Einstein of adapting our modes
of perception borrowed from the sensations to the gradually deepening knowledge of
the laws of nature. The hindrances met on this path originate above all in the fact that,
so to say, every word in the language refers to our ordinary perception.
where c' is the velocity of light measured on the Earth (assumed to be an inertial
reference system), c is the velocity of light in the free ether as measured by someone
in a reference system fixed in the ether, and v is the relative velocity between the
Earth and ether. Eqs. (3) and (4) are inconsistent with each other. Devilishly clever
hypotheses were invented to explain away disagreement between data and theory.
Amongst them was the Lorentz Contraction Hypothesis. Essentially physicists took
v to be a causative agent that, for example, caused a body to contract in the direction
of its motion, owing to the interaction between the moving body's constituent
40 ARTHUR 1. MILLER
electrons and the ether. And these hypotheses were pretty much acceptable to most
everyone, until special relativity was finally understood in 1911. Einstein removed
the inconsistency between Eqs. (3) and (4) by declaring Eq. (3) incorrect and Eq. (4)
precise both theoretically and experimentally. This bold move was among the
avenues that led him to discover the relativity of time, which seems at first to be
inconsistent with our Galilean-Newtonian common sense, that is, intuition.
4. CONCLUDING COMMENTS
What are we to make of all this? Amongst the insights that emerge is that science
can progress despite, and sometimes because of, inconsistencies lurking in
reasoning. Although this is nothing new to historians of science, framing it in terms
of sources of inconsistencies brings the concept of scientific progress into clearer
focus.
What is deeply interesting, I believe, is why approximations work and how they
bear on scientific progress. Briefly: as to why they work, I think the best we can
presently say is that we are lucky to live in a benign part of the universe. And I do
not want this to be construed as any version of an anthropic principle. I mean that
the Earth can be considered as an inertial reference system, and force fields are such
that we can cut Nature at her joints and pretty much treat the four known forces
separately. Galileo, for example, was able to consider the Earth as an inertial
reference system and treat pendula as if they were undergoing simple harmonic
motion.
We have noticed that simple fluids were assumed at first in basic research in heat
and electromagnetic theory in order to conform to our intuition of the way
phenomena ought to occur, as we saw with caloric, electrical fluids and the ether.
Although all of these concepts were jettisoned they pointed to fruitful ways to
proceed. We may say that, in the end, none of these concepts referred. Nor for that
matter may the concept of the fundamental electron, if new data from the particle
accelerator in Hamburg is borne out. Nevertheless, we need not fear what Hilary
Putnam calls the disastrous "meta-induction" where, in the end, nothing refers and
science is merely some sort of empiricist game wherein scientific theories are just
economic ways of collating experimental data (Putnam 1978, 25). In no uncertain
terms the history of science is testimony to the resounding continuity of theoretical
structures (e.g., see Miller 1996, esp. Chapter 7).
In this way so, too, are our modes of intuition transformed by emergent scientific
theories. For example, the apparent inconsistency or paradox between Eqs. (1) and
(2) is removed by realising that the concepts "wave" and "particle" from the world
of sense perceptions no longer apply to atomic phenomena; and our notion of the
absoluteness of space and time from Newtonian science becomes transformed
through removal of the inconsistency between Eqs. (3) and (4). The theories that
emerge from one another are better and better approximations to an underlying
physical reality that lay beyond sense perceptions.
REFERENCES
Bohr, N. (1921), L'Application de la theorie des quanta aux problemes atomiques. In A tomes et
Electrons, Paris: Gauthier-Villars, pp. 228-247. (This is the proceedings of the 1921 Solvay
Conference.)
Bohr, N. (1928), The Quantum Postulate and the Recent Development of Atomic Theory. Nature
(Supplement), 580-590. (This is a published version of Bohr's lecture delivered on 16 September
1927 to the International Congress of Physics, Como, Italy.)
Einstein, A. (1923), Fundamental Ideas and Problems of the Theory of Relativity. Nobel Lectures,
Physics: 1901-1921, New York: Elsevier, pp. 479-490.
Galilei, G. (161311957), Letters on Sunspots. In Discoveries and Opinions ofGalileo, S. Drake (ed. and
trans later), New York: Doubleday.
Galilei, G. (163211967), Dialogue Concerning the Two Chief World Systems-Ptolemaic & Copernican.
Berkeley: University of Cali fomi a Press (translated by S. Drake).
Michelson, A. A. and E. W. Morley (1887), On the Relative Motion of the Earth and Luminiferous Ether.
American Journal of Science XXXIV, 333-345.
Miller, A. I. (1981), Albert Einstein's Special Theory of Relativity: Emergence (1905) and Early
Interpretation (1905-1911). Reading, MA: Addison-Wesley. Reprinted in 1998 by New York:
Springer-Verlag.
Miller, A. I. (1986), Imagery in Scientific Thought: Creating 20th-Century Physics. Cambridge: MIT
Press.
Miller, A.1. (1992), Imagery, Probability and the Roots of Werner Heisenberg's Uncertainty Principle
Paper. In Sixty- Two Years of Uncertainty: Historical, Philosophical, and Physical Inquiries into the
Foundations of Quantum Mechanics, A. I. Miller (ed.), New York: Plenum, pp. 3-15.
Miller, A.1. (1994), Early Quantum Electrodynamics: A Source Book. Cambridge: Cambridge University
Press.
Miller, A. I. (1996), Insights of Genius: Imagery and Creativity in Science and Art. New York: Springer-
Verlag.
Planck, M. (1910), Zur Theorie der Warmestrahlung, Annalen der Physik 31,758-767.
Poincare, H. (1900), La theorie de Lorentz et Ie principe de reaction. In Recueil de travaux ofJerts par les
auteurs aH.A. Lorentz, The Hague: Nijhoff, pp. 252-278.
Putnam, H. (1978), Meaning and the Moral Sciences. Boston: Routledge & Kegan Paul.
Richardson, O. W. (1916), The Electron Theory of Matter. Cambridge: Cambridge University Press.
JEAN PAUL V AN BENDEGEM
1. INTRODUCTION
In this paper I will not confine myself exclusively to historical considerations. Both
philosophical and technical matters will be raised, all with the purpose of trying to
understand (better) what Newton, Leibniz and the many precursors (might have)
meant when they talked about infinitesimals. The technical part will consist of an
analysis why apparently infinitesimals have resisted so well to be formally
expressed. The philosophical part, actually the most important part of this paper,
concerns a discussion that has been going on for some decennia now. After the
Kuhnian revolution in philosophy of science, notwithstanding Kuhn's own
suggestion that mathematics is something quite special, the question was
nevertheless asked how mathematics develops. Are there revolutions in
mathematics? If so, what do we have to think of? If not, why do they not occur? Is
mathematics the so often claimed totally free creation of the human spirit? As usual,
there is a continuum of positions, but let me sketch briefly the two extremes: the
compietists (as I call them) on the one hand, and the contingents (as I call them as
well) on the other hand.
A completist basically defends the thesis that mathematics has a definite, if not a
"forced" route to follow in its development. As a matter of fact, mathematics strives
for its own completion. In that sense, there can be no revolutions, because the
outcome of such a revolution is not settled beforehand. Whereas in the completist's
case, there is a definite direction. Usually, this implies a (large-scale») linear growth
and a cumulative growth of mathematical knowledge. The deeper philosophical
* My most sincere thanks to loke Meheus and Diderik Batens for thoroughly reading the paper and for
indicating some grave errors that, I hope, have been set right in this new version. Without going into
too much detail, apparently, although I reject the completist's option, unconsciously I was still
thinking and acting as one. Thanks also to Bart Van Kerkhove for helpful suggestions and remarks.
) I emphasize large-scale because one cannot exclude human error in the short term. It is very instructive
to have a look at, e.g., Altmann's analysis of the development of quatemions in the work of Hamilton
(1992, chapter 2). Not exactly a comedy of errors, but rather an amusing play how to arrive at the
right notion using the wrong picture.
43
J. Meheus (ed.), Inconsistency in Science, 43-57.
© 2002 Kluwer Academic Publishers.
44 JEAN P AUL VAN BENDEGEM
Why do I believe that these philosophical considerations are so important for (an
understanding of) the history of infinitesimal calculus? For the simple reason that
from the completist's point of view, infinitesimal calculus, as conceived before and
2 It is not my intention to present a full overview of the contingents. As a first guide, the reader can
consult Restivo et al. 1993, Gillies 1992 or Hersh 1997.
INCONSISTENCIES IN THE HISTORY OF MATHEMATICS 45
Logic", p. 376, they present a very simple, yet devastating argument that shows that
a naive formalisation of (inconsistent) infinitesimals leads to triviality (or to
something quite close). In that sense it pleads against the use, if not the existence of
infinitesimals. I will therefore call this counterargumene A:
Suppose that we have a theory of infinitesimals such that we both have an axiom
stating that infinitesimals are equal to zero-dx = O--and an axiom stating that
the opposite-dx *' O-holds. If the usual rules for the real numbers apply, the
following argument can be built up:
Suppose we have two sets of numbers R = {r, r', ... } (standard real numbers)
and Inf = {E, 1::', ... } (specific infinitesimals), such that R n Inf = 0. Suppose
further that arithmetical operations, such as addition and subtraction, are defined
over R U Inf. On the one hand, we do want that for an arbitrary real number r, it
is the case that for an arbitrary infinitesimal 1::, r *' 1::. But it is easy to show that
the contrary statement holds as well. It is sufficient to ask the question whether
r + I:: E R or r + I:: E Inf, for any pair rand I::?
(B I) If r + I:: E R, then (r + 1::) - r E R, hence I:: E R, thus every infinitesi-
mal equals some real number.
(B2) If r + I:: E Inf, then (r + 1::) - I:: E Inf, hence r E Inf, thus every real
number equals some infinitesimal.
(B3) It follows that the distinction between Rand Inf is completely lost,
contradicting (at least) R n Inf = 0.
When we look (even briefly) at the historical material, e.g., Boyer 1959, then it is
clear that the same questions bothered everybody at the time:
3 An important distinction must be made here. It is a counterargument for the view that in present-day
mathematics infinitesimals can still (and perhaps should) play an important part as they are easier and
simplier to handle than limit operations. It is not a counterargument for the historical problem: what is
it that Newton, Leibniz and others had in mind, when they used infinitesimals? Here we can only
conclude that whatever it is, it must be a complex thing, which is actually what Priest and Routley
claim.
INCONSISTENCIES IN THE HISTORY OF MATHEMATICS 47
In this paragraph I will outline how one can have "genuine" infinitesimals on
condition that one is willing to accept the following:
(a) in terms of models, only local models (in a sense to be specified in what follows)
are considered, or, alternatively, there are no global models 4 ,
(b) all local models are essentially finite.
I realize that these conditions run counter to almost anything that is cherished by
logicians, mathematicians and philosophers. I will return to this point in paragraph 6.
If, however, one is willing to make these "sacrifices", then matters become rather
easy, if somewhat tedious. What follows presents a rough outline and not a full-
blown theory. Let us start with the standard theory T of real numbers. The first
change that has to be made is that two sets of distinct variables will be used:
(i) variables for "standard" real numbers: x, y, z, ...
(ii) variables for "infinitesimals": E, E', E", ...
Suppose we now have a finite series of formulas F = {F" F 2, ..• , Fn}, all expressed
in the language of T. The intuitive idea is that F could, e.g., represent a calculation
of the value of a function in a particular point. Further suppose that if all formulas
are interpreted in R such that all variables are treated in the same way, then they are
true in the standard model of real numbers.
Obviously, if F is finite, then so are the number of variables, both "standard" and
"infinitesimal", so is the number of constants, and so is the set of terms occurring in
the members ofF.
(tl) some tenns involve only constants and variables for standard real
numbers
(t2) some terms involve only infinitesimal variables
(t3) some tenns are mixed such that the tenn consists of the sum of a tenn
of type (tl) and a mixed tenn
(t4) some terms are mixed such that the tenn consists of the product of a
tenn of type (t2) and a mixed tenn
I will make one further assumption, namely, that, although (tl)-(t4) do not exhaust
the set of tenns, yet, any tenn can be transfonned into one of these categories. 5
Thus, e.g., if the tenn is (x + e)' e + e' (x - e), it is mixed but neither of type (t3)
nor type (t4). But the tenn is easily transfonnable into e' (x + e + x - e), and thus of
type (t4).
(S I) Let Int fix the values of the standard variables and of the constants.
This also implies that all terms of type (tl) are thereby fixed, as we
follow standard procedures, i.e., Int(tl + t2) = Int(t 1) EEl Int(t2), where
EEl is addition in the real number system, likewise for multiplication,
(S2) Consider the following set Dist = {lInt(t 1) - Int(t2)1, IInt(t3)I I tl, t2
are tenns of type (tl) and t3 is the tenn that has the smallest non-zero
absolute value}. In short the set Dist looks at all the differences
5 A simple proof by induction on the length of terms will do. Suppose that all terms of length n are of type
(tl)-(t4). A term I of length n + 1 is either of the form I' + I" or I'· t". As both t' and t" are
shorter than n, they are of type (tl)-(t4).lt is now sufficient to check all possible cases to see that tis
of either one of the four types. E.g., suppose I' is (t3) and I" is (t4), then t' is of the form tl + m,
where II is of type (tl) and m is mixed, and I" is of the form I"m', where I, is of type (t2). But then
1= t' + I" = II + m + I,' m' = II + m", where m" is obviously a mixed term.
INCONSISTENCIES IN THE HISTORY OF MATHEMATICS 49
Example (continued):
Suppose that Int(x) == 2. Then terms of type (tl) are evaluated as:
Int(x2) == 4, Int(x3) == 8, Int(3) == 3, Int(3x) == 6, Int(3x2) == 24.
The minimum distance is 1, so take, e.g., 8' == 0.00l.
The largest term of type (t4) we can encounter is 3X2E + 3XE2 + E3.
A direct calculation shows that Int(E) = 0.00001 will do the job.
Graphically, what is being proposed here, produces the following picture. The
line represents the real numbers R, where only a finite number of elements are
interpreted, such as x, E, and so on.
[ -d, dJ .. .[Int(x)-d, Int(x) E!J dJ·· .[Int(x2) -d, Int(r) E!J dJ·· .[Int(~)-d, Int(~) E!J dJ·.·
R
o Int(x)
Furthermore one has to make sure that in the interval [- d, dJ one finds the
interpretations of all infinitesimal expressions, such as, E, E2, E\ ... and that for any
term t of type (tl), all expressions t + E' t' (where t' is any term) are interpreted in
the interval [Int (t) - d, Int(t) E!J d]. Because we deal with only a finite number of
statements, this procedure can always be executed.
On the one hand, it is obvious that I treat infinitesimals, such as E in the example,
as an ordinary real number and, semantically, it is in fact interpreted as such. But, on
the other hand, it is also the case that, in every model MP-model of a set of formulae
F, for any "standard" variable x and for any infinitesimal variable E, x =1= E, thus
expressing that no "standard" number equals an infinitesimal. Likewise, for all the
constants n named in the set F, n =1= E. Thus infinitesimals are at the same time
different.
F\: 0+0=0
F2: E+ E= E
F3: 2· E = E
F4 : 2= I
Terms of type (tl) are the constants 0, I and 2. If these are given their standard
interpretation, then 0 = 1. The terms of type (t2) are E, E + E, 2· E. Take for
Int(E) = 0.01. Then (I) will turn out true, but (2) will not. Moreover, it is not
difficult to see that whatever value is taken for E, since the statement E = 0 is false,
the reasoning is blocked.
Solving counterargument B is equally simple and straightforward. Because of the
finite number of terms, although they are interpreted in the same domain, a
distinction can be made between real ("standard") numbers and infinitesimals. Thus
syntactically, for a particular set F of formulas, as said above, one can claim that
x '* E, yet, semantically, both are real numbers.
(b) Perhaps the most important point to note is that, for a given set F, there does not
necessarily exist a MP-model, although a classical model does exist. A simple
example will illustrate this. Consider the simple set F = {(3E)(E + I = 2),
(3x)(x + I = I)}. In a classical model, if Int(E) = 1 and Int(x) = 0, then both
formulas are true. But according to the construction for a MP-model, one first sets
Int(x) = 0, Int(l) = 1, Int(2) = 2, Int(x + 1) = Int(x) EB 1 = 1. Dist = {O, 1, 2}, thus
0= 1. For 0' and for E choose 0.001, then Int(E + 1) = 1.001. This means that the
first formula of F can never be true. Hence there is no MP-model.
(c) The restrictions mentioned in (a) also have an effect on the underlying logic.
Take a simple example, suppose one has a statement such as (''liE)(t(E) = t'(E»,
where t and t' are terms of type (t2). Suppose further that in the set F to which this
formula belongs, a finite number of constants are mentioned. Take such a constant n,
then it does not follow from (VE)(t(E) = t'(E» that ten) = t'(n). In the standard real
number model this will be no problem, but if the formula ten) = t'(n) is added to F,
then the MP-model might very well have to be modified, for we now have an
additional formula involving a constant. It therefore becomes a very delicate process
to see what formulas can or cannot be added to a given set F. In one sentence: given
INCONSISTENCIES IN THE HISTORY OF MATHEMATICS 51
a MP-model for a given set F, then this model will not necessarily be aMP-model
for F U {logical consequences of F}.
Similar comments can be made about substitution of identicals. Suppose that x
occurs in some formula ofF, that states that x = 1. Now, obviously, 1 = n' (lIn), but
it does not follow necessarily that the formula x = n . (lIn) can be added to F, for we
have introduced two new constants to evaluate, namely, n and lin.
(d) To continue the observation made in (c), consider the following situation.
Suppose that we have the set F as in the example presented above. In at least one of
the MP-models, the interpretation of the variable E will be Int(l':) = 0.00001. Hence,
if we assume for simplicity that 0.00001 is syntactically represented as 0.00001,
then the formula (3E)(1': = 0.00001) will be true. Hence, this formula belongs to the
consequences of F. Suppose that we now add explicitly this formula. Then 0.0000 I
behaves as a new constant, thus we will have to recalculate the set Dist, which will
lead to a new value of 0, hence to a new value of 0' and hence to a new value of E.
Int(l':) will inevitably be smaller than 0.00001, as this constant has now become a
"standard" number. But then the statement (31':)(1': = 0.00001) ceases to be true. In
fact, no such statement of this form can be added, for it puts an infinitesimal equal to
a constant. Or, which amounts to the same: from (3E)(1': = 0.00001) follows that
(3x)(3E)(E = x) and that has to be excluded. All this means that, although I': has a
specific value in the model for a set of formulae F, this is not expressible in terms of
a formula that could be added to F.
(e) Are there formulas that one can safely add to a given set F? One such example
has been mentioned already: namely that for any specific term t of type (tl), it will
be the case that
~(3E)(t = E)
Actually, a stronger statement can be obtained: for any specific term t of type (tl),
(VI':)((t> 0) J (E < t»
In contradistinction to non-standard analysis we do not have statements such as:
As it should be, for otherwise we would have "absolute" infinitesimals, which is not
the case.
Example:
(I) «x + g)3 - x 3)/g == «x 3 + 3x2 g + 3xg 2 + g3) - x 3)/g
(2) «X3 + 3x2 g + 3xg 2 + g3) - x 3)/g == (3X2g + 3xg2 + g3)/g
(3) (3X2g + 3xg2 + g3)/g == 3x2 + 3xg + g2
(4) 3x2 + 3xg + g2 == 3x2 + (3x + g). g
(5) «x + g)3 -x3)/g == 3x2 + (3x + g). g
Together with g == 0, the equations become respectively:
(I) Df == 0/0
(2) % == 0/0
(3) % = 3x 2
(4) 3x 2 == 3x 2
(5) Df == 3x 2
What remains as meaningful expression are (4) and (5), and the latter one gives
us the answer
Let me first of all present some arguments to show that MP is rather close to what
mathematicians of the Newton-Leibniz era (possibly) had in mind. If I can rely on
the analysis of Hide Ishiguro in his 1990, then it seems clear that for Leibniz
INCONSISTENCIES IN THE HISTORY OF MATHEMATICS 53
infinitesimals do not exist. The following quotes, taken from Ishiguro's analysis,
illustrate this view (page numbers refer to Ishiguro 1990):
... every number is finite and specifiable, every line is also finite and specifiable. Infinite
[magnitudes) and the infinitely small only signify magnitudes which one can take as big
or as small as one wishes, in order to show that the error is smaller than the one that has
been specified. [from the Theodicy, p. 79)
and
As concerns infinitesimal terms, it seems not only that we never get to such terms, but
that there are none in nature, that is, that they are not possible. Otherwise, as I have
already said, I admit that if I could concede their possibility, I should concede their
being. [letter to Johann Bernoulli, p. 84)
and, finally,
Meanwhile I have shown that these expressions are of great use for the abbreviation of
thought and thus for discovery as they cannot lead to error, since it is sufficient to
substitute for the infinitely small, as small a thing as one may wish, so that the error
may be less than any given amount, hence it follows that there can be no error. [letter to
Des Bosses, p. 85)
The basic idea that comes forward is precisely the idea that, although we use the
word "infinitesimal", it does not follow that this word must refer to or designate
something "special". This was the guiding principle for the development of MP. If
distinct terms do not necessarily need distinct designations-both "reals" and
"infinitesimals" are interpreted as standard reals-then the connection between
syntax and semantics cannot be as close as we usually want it to be. For all clarity, I
must repeat a remark made earlier: MP only serves to sharpen our historical ideas. It
is anybody's guess whether Leibniz would have been pleased with the idea of local,
finite models that cannot generate one global model just for the purpose of being
able to talk about infinitesimals. Probably not.
If MP does provide a possibly historically plausible reconstruction of infinitesi-
mals, then, of course, the answer to the question: "Why did analysis develop in an
entirely different direction leading to the elimination of infinitesimals altogether?",
becomes quite simple. The answer has been given above. It is the fact that the
statement
is not the case in MP, i.e., cannot be added to any finite set F of formulas, whereas
the statement
can be added to any such set. A "free" translation of this observation is that
infinitesimals are local entities, whereas the former confirms the existence of global
infinitesimals. Therefore, as mathematicians have taken an option for the global
model, then, inevitably, infinitesimals had to go, and the E-o-analysis became an
attractive alternative, as it did promise global models. In this sense, the notion of
54 JEAN PAUL V AN BENDEGEM
"infinitesimal" has shared the same faith as some vague predicates that were used
once in mathematics, e.g., "small" and "large" numbers. I will not go into the details
here (see Van Bendegem 2000), but, making use of the supervaluation approach to
vague predicates, it is possible to restore (part of) the use of such predicates. Just as
in the case of MP, it becomes possible to reason in a rigorous fashion about them
and, hence, to prove such theorems as "Small numbers have a few prime factors".
Of course, one might argue from a completist's perspective that global models
are to be preferred over a set of local models and there might very well be excellent
arguments to support that view, but, at the same time, it shows that the option to go
for the global was neither necessary nor "forced" in one way or another. 7
As the label indicates, in contrast with NSA where infinitesimals are introduced
arithmetically and/or algebraically, in SDG infinitesimals are geometric entities. In
so few words, the basic idea of SDG is to think the world in a continuous way. A
curve in the plane, e.g., is locally straight and putting all these straight bits together
produces the curve. In number language, such a straight piece is represented by a
number E, such that E '*' 0, yet E2 = 0. Such numbers E, called minims or non-
degenerated linelets form a set M and that is to be the set of infinitesimals.
Of course, it is obvious that within the real number system such a set cannot
exist. Hence, the question is whether such a structure is actually possible. As it
happens, within the framework of category theory, more specifically, the theory of
toposes, it is possible to show that such structures exist. It follows, in addition, that
one also has to change the underlying logic. Classical logic will not do, one has to
move to intuitionistic logic, as, e.g., the principle of the excluded third (either p or
not p) fails. This is, up to my knowledge, the only one of the four alternatives
presented in this paragraph, that forces one to change logic. As John Bell remarks:
"This result is remarkable insofar as it shows that a purely mathematical postulate-
the Principle-can have a purely logical consequence-the failure of the law of
excluded middle." (1988, 294). The principle referred to is the principle that
expresses the local linearity of functions: for a function f(x), there is a unique
number a such thatf(x + E) = f(x) + a' E.
Oddly enough, Leibniz appears in this context too but from a different viewpoint.
It does not concern Leibniz's idea of infinitesimals directly, but rather his prinCiple of
continuity, asserting that all natural processes occur continuously. Indeed, it is possible
to show in SDG that all functions are continuous. As is the case for NSA, the question
remains however how close this approach is to Leibniz's ideas.
This approach suggested by Donald Knuth is very close to (at least the aims of)
MP. 9 As he writes:
8 A rather interesting historical datum: Georg Cantor, the founding father of transfinite ordinal and
cardinal numbers-one of the basic ingredients to generate non-standard models and thus to lead to
non-standard analysis-has always rejected the notion of infinitesimals as intrinsically contradictory.
Rejection is actually a mild expression as, e.g., in a letter to Giulio Vivanti Cantor he remarked that
Johannes Thomae was the first to infect mathematics with the Cholera-Bacillus of infinitesimals (see
Dauben 1979, 131). So one has every reason to assume that Cantor would not have welcomed non-
standard analysis.
9 What follows is based on Knuth 1998. This paper, I should mention, is Uust) a letter to the editor. More
is to be found (but not dealt with in this paper) at http://www-cs-faculty.stanford.edu/-knuth! calc/.
56 JEAN PAUL V AN BENDEGEM
This notation, first used by Bachmann in 1894 and later popularized by Landau, has the
great virtue that it makes calculations simpler, so it simplifies many parts of the subject,
yet it is highly intuitive and easily learned. [Knuth 1998, 687]
The basic idea is, first, to introduce a new function A(x), its intuitive meaning being
"absolutely at most x". It is easy to show that one has nice properties such as
x = A(x) or 10A(Z) = A(lOO). There is a drawback: equality is affected. From
3 = A(5) and 4 = A(5) does not follow that 3 = 4. With the function A, a new
function is introduced, namely O(x), meaning "the x such that x is equal to e· A(x),
for some non-specified constant e". With this function, it becomes easy to define a
"strong" derivative (in Knuth's terms). It is the functionf'(x) such that:
for £ sufficiently small. Do note, however, that in contrast to MP, the fact that the
symbol £ is used, does not mean that it is to be thought distinct from x. The only
requirement is that £ is sufficiently small, so any real variable will do.
This approach has been developed by J.M. Henle in his 1999. A similarity between
his view and MP is that infinitesimals are not separate entities but made up from real
numbers. In fact, in his construction infinitesimals are simply sequences of reals
tending to O. The reals themselves can be easily recovered as constant sequences,
i.e., with the real number r corresponds the sequence (r, r, r, ... ). The cleverness of
NNSA is that by speaking directly about the sequences-that is, by having names
for them~a lot of "quantifier talk" is eliminated (or buried in the background, if you
like). If, e.g., a is a sequence, then a > 2 means: for some k, for all n, if n ~ k, then
an> 2, where a = (a" az, ... , ak, ... ). The resulting calculus is surprisingly close to
an informal way of thinking about infinitesimals. There is one feature that is worth
mentioning in this brief presentation: the classical reals are completely ordered, but
for sequences this is not the case. To use Henle's example: the sequences
A = (1,0,1,0, ... ) and B = (0, 1,0,1, ... ) are incomparable. This means that in a
curious way the sequences behave, at least in this respect, more intuitionistically
than classically.
As a closing remark, let me emphasize that the fact that these proposals are
mentioned here, does not imply that they are also involved in the completist-
contingent debate. However, to repeat, they show that alternatives exist and do so
abundantly.
REFERENCES
Altmann, S. L. (1992), Icons and Symmetries. Oxford: Clarendon Press.
Bell, 1. L. (1986), From Absolute to Local Mathematics. Synthese 69,409-426.
Bell, 1. L. (1988), Infinitesimals. Synthese 75, 285-315.
INCONSISTENCIES IN THE HISTORY OF MATHEMATICS 57
Bell, J. L. (1995), Infinitesimals and the Continuum. The Mathematical Intelligencer 17, 55-57.
Bloor, D. (1976), Knowledge and Social Imagery. London: RKP. Second edition: Chicago: University of
Chicago Press, 1991.
Boyer, C. B. (1959), The History of the Calculus and its Conceptual Development. New York: Dover
Books.
Dauben, J. W. (1979), Georg Cantor. His Mathematics and Philosophy of the Infinite. Cambridge,
Mass.: Harvard University Press.
Dauben, J. W. (1995), Abraham Robinson. The Creation of Nonstandard Analysis. A Personal and
Mathematical Odyssey. Princeton: Princeton University Press.
Gillies, D. (1992), (ed.), Revolutions in Mathematics. Oxford: Clarendon Press.
Grattan-Guinness, I. (1992), (ed.), Encyclopedia of the History and Philosophy of the Mathematical
Sciences. London: Routledge.
Grattan-Guinness, I. (1997), The Fontana History of the Mathematical Sciences. London: Fontana Press.
Henle, J. M. (1999), Non-nonstandard Analysis: Real Infinitesimals. The Mathematical Intelligencer 21,
67-73.
Hersh, R. (1997), What is Mathematics, Really? London: Jonathan Cape.
Ishiguro, H. (1990), Leibniz 's Philosophy of Logic and Language. Cambridge: Cambridge University
Press.
Knuth, D. (1998), Teach Calculus with Big 0 (Letter to the Editor), Notices of the AMS 45, 687-688.
Lakatos, I. (1976), Proofs and Refutations. Cambridge: Cambridge University Press.
MacLane, S. (1986), Mathematics: Form and Function. Heidelberg: Springer-Verlag.
McLarty, C. (1992), Elementary Categories, Elementary Toposes. Oxford: Oxford University Press.
Moerdijk, I. and G. E. Reyes (1991), Models for Smooth Infinitesimal Analysis. Heidelberg: Springer-
Verlag.
Naisse, J. P. (1992), L 'approximation analytique. Vers une tMorie empirique, constructive etfinie.
Brussels: Editions de I'Universite de Bruxelles.
Newton, I. (1962), (Motte's Translation. Revised by Cajori), Principia. Vol. I: The Motion of Bodies. Vol.
II: The System of the World. Berkeley: University of California Press.
Priest, G., R. Routley and J. Norman (eds.) (1989), Paraconsistent Logic. Essays on the Inconsistent.
Mlinchen: Philosophia Verlag.
Restivo, S., J. P. Van Bendegem and R. Fischer (eds.), (1993), Math Worlds: New Directions in the
Social Studies and Philosophy of Mathematics. New York: State University New York Press.
Van Bendegem, J. P. (1993), Real-Life Mathematics versus Ideal Mathematics: The Ugly Truth. In
Empirical Logic and Public Debate. Essays in Honour of Else M. Barth. E. C. W. Krabbe, R. J.
Dalitz and P. A. Smit (eds.), 1993, Amsterdam: Rodopi, pp. 263-272.
Van Bendegem, J. P. (1999), The Creative Growth of Mathematics. Philosophica 63, 119-152.
Van Bendegem, J. P. (2000), Alternative Mathematics: The Vague Way. In Festschrift in Honour of
Newton C. A. da Costa on the Occasion of his Seventieth Birthday. D. Krause, S. French and F. A.
Doria (eds.), Synthese 125, pp. 19-31.
OTAVIa BUENO
1. INTRODUCTION
* Many thanks to Newton da Costa, Steven French, Joke Meheus, and Sarah Kattau for helpful
discussions on the issues examined here.
59
J. Meheus (ed.), Inconsistency in Science, 59-79.
© 2002 Kluwer Academic Publishers.
60 OTAVIO BUENO
In my view, there are at least two different reactions to this position. The first
stresses that once appropriate patterns of mathematical change are spelled out, all the
apparent inconsistencies among mathematical theories are dissolved-one should
ultimately find consistent formulations of these theories. Thus, once these
formulations are found, we are entitled to learn from the study of the dynamics of
mathematics. Of course, this reaction adopts a fairly conservative view towards
inconsistencies, and basically assumes that all reliable information should be, at
least, consistent.
The second reaction will challenge this assumption. The existence of inconsis-
tencies within bodies of mathematical information does not necessarily preclude us
from learning from such bodies. In particular, even if proposed patterns of
mathematical change involve inconsistencies, we cannot conclude from this that
such patterns are hopelessly unreliable as a way of providing information about the
nature of mathematics. Notice that we have here, in fact, two different levels of
inconsistency. The first is concerned with inconsistencies at the theoretical level,
involving particular mathematical theories; the second concerns inconsistencies at
the meta theoretical level, and this involves our representations of mathematical
change. What the second reaction countenances is a unified strategy of
accommodating inconsistencies-at face value-at both levels.
The constraint of taking inconsistencies at face value is important. The idea is
not to claim that, once mathematical theories are appropriately reformulated, those
inconsistencies can either be shown to be innocuous, or simply are dissolved. This
would simply turn the second reaction into a variation of the first. As opposed to
this, the second reaction urges one to find an appropriate epistemic role for
inconsistencies. Once we have an appropriate framework in terms of which
mathematical change can be modelled, the second reaction will also have to provide
a clear strategy that allows us to accommodate inconsistencies in a more 'positive'
way, i.e. that assigns a positive role for inconsistencies within mathematics and
mathematical change. The existence of important mathematical theories that are
inconsistent (such as the calculus and naive set theory) is enough to demand such a
VIew.
In this paper, I am concerned with articulating and defending this second
reaction. In order to do so, I shall first put forward an account of what sort of
epistemic unit should be taken as basic. As we shall see in sections 2 and 3, I take it
that two independent proposals found in Kitcher 1984 and Laudan 1984, if
adequately combined, provide the first step towards an appropriate framework to do
so. This move, however, is not sufficient as a defence of the second reaction, since
the resulting framework cannot accommodate inconsistencies. In order to pursue this
task-which is the chief aim of section 3-1 propose that we adopt the conceptual
resources provided by da Costa and French's partial structures approach (see, for
instance, da Costa and French 1989, 1990, 1993a, and 1995). As we shall see in
section 3.1, three important features are brought by this view: a broader notion of
structure (partial structure), a weaker notion of truth (quasi-truth), and a
paraconsistent setting (in terms of the underlying logic of quasi-truth: a convenient
version of laskowski's logic; see da Costa, Bueno and French 1998). In this way, or
so I shall argue in section 3.2, we can put forward an account of inconsistencies
MA THEMATICAL CHANGE AND INCONSISTENCY 61
As is well known, there is an old tradition of interpretation of science that adopts the
strategy of putting forward a view about the ways in which science evolves in order
to draw general conclusions concerning the nature of scientific knowledge. This
tradition takes the notion of scientific change as basic for the understanding of the
nature of science and the knowledge it supplies. Of course, this is not to say that
problems unrelated to the dynamics of scientific knowledge are not to be considered
within such a view; they are. But the analysis proposed is (to be) articulated in terms
of an account of scientific change. Indeed, it is at this level that the 'explanatory
power' of this tradition can be found.
Kitcher's account of the nature of mathematical knowledge (as presented in
Kitcher 1984) certainly belongs to this tradition. The role that Kitcher assigns to
scientific change-in particular, to mathematical change-parallels in importance
that found within, for instance, either Popper's or Lakatos's proposals: the task of
taking into account scientific (or mathematical) knowledge is to be achieved mainly
by the formulation of a theory of scientific (or mathematical) change. l
However, Kitcher's view of mathematical change is embedded in a general
epistemological framework, and much of the motivation for his proposals rests on
his critical appraisal of previous epistemological accounts of mathematics. The
relevance of mathematical change is presented in the context of a critique of two
interrelated doctrines that have characterised such previous accounts, namely,
mathematical apriorism (the claim that mathematical knowledge is a priori), and
mathematical 'individualism' (a lack of consideration of the role of mathematical
community in the development of mathematics). As opposed to the former, Kitcher
puts forward his mathematical empiricism, rejecting earlier empiricist interpretations
of mathematics, such as Lakatos's, Putnam's and Quine's (see Kitcher 1984,4). As
opposed to the latter, he advances a theory of mathematical practice, motivated by
Kuhn's view, but stripped of his general framework/ in order to claim that the
development of mathematical knowledge occurs in terms of a 'rational modification
1 Actually, one of Kitcher's main points consists in spelling out the similarities (without disregarding the
differences) between scientific and mathematical changes, in order to motivate his account of the
methodology of mathematics (see Kitcher 1984, 150, and the rest of Chapter 7). Indeed, one of his
claims is that 'the growth of mathematical knowledge is far more similar to the growth of scientific
knowledge than is usually appreciated' (Kitcher 1984, 8). Of course, such a claim clearly echoes
Lakatos's view about the connection between scientific and mathematical knowledge (for details, see
Lakatos 1976 and Lakatos 1978b).
2 In particular, the Kuhnian notion of paradigm is not accepted (see Kitcher 1984, 163-164).
62 OTAvIOBuENO
of mathematical practices' (Kitcher 1984, 165). These practices, which are the basic
epistemic unit of mathematical change, have five components: a language, certain
accepted (mathematical) statements, certain accepted types of reasoning (an
underlying logic), certain questions considered as important and certain
metamathematical standards with regard to proofs, definitions and so on (Kitcher
1984, 163). The introduction of this notion of mathematical practice is Kitcher's
way of assigning a function to the mathematical community within his proposal,
even though in a rather idealised way.
As a result of the shift to mathematical empiricism and the emphasis on
mathematical practice, Kitcher's account of mathematics acquires a historicist
outlook: mathematical knowledge is obtained by extending the knowledge produced
by previous generations (see Kitcher 1984, 4-5). At this point, it is natural to ask,
working backwards in the chain of generations, how the knowledge of the first
generations was produced at all? To this question, Kitcher formulates a surprising
answer: he traces back this knowledge to 'ordinary perception' (1984,5). In order to
explain the possibility of obtaining mathematical knowledge from perception,
Kitcher devises a theory of mathematical reality, of 'what mathematics is about'
(1984, 10 1-148). In order then to show how from such perceptual origins
mathematics can be developed into such an impressive body of theories, techniques
and results that one finds today, he puts forward his account of the growth of
mathematical knowledge (1984,149-228). These two doctrines constitute the core of
Kitcher's mathematical empiricism.
Thus, Kitcher's account of mathematical change has two important features.
Firstly, it represents mathematical change in terms of broad epistemic units,
mathematical practices (in Kitcher's sense). Secondly, because of this feature, it
allows one to formally accommodate the role of mathematical activity and practice
in the changes of mathematical theories. As a result, so Kitcher argues, a more
faithful account of the dynamics of mathematics, vis-a-vis the historical record, is
put forward.
In this picture, each particular historical moment is characterised by a given
mathematical practice, involving a community of mathematicians who adopt the
same language to investigate a given mathematical domain, who share a common
core of accepted mathematical statements and open questions, and who, in trying to
settle the latter, adopt the same metamathematical standards and the same logic.
Mathematical change results from the interplay of each of these five components, as
well as changes in them. For instance, new, more demanding patterns of rigour (a
change in metamathematical standards) may lead to the rejection of a previously
accepted proof and to the demand for a new proof of a given result. Similarly, the
rejection of a previously accepted mathematical statement S will bring an epistemic
change in all results which depend on S, requiring new proofs of them independent
of S. Moreover, by considering some open questions as no longer important,
changes are introduced in the way mathematics is practised. This may result from
the acceptance of new mathematical statements, which change the relative
importance of the open questions, since they change the direction of the research in
a particular mathematical domain. Furthermore, by changing the underlying logic of
a given mathematical practice, new inferential patterns that have been previously
MATHEMATICAL CHANGE AND INCONSISTENCY 63
The partial structures approach relies on three main notions: partial relation, partial
structure and quasi-truth. 3 One of the main motivations for introducing this proposal
comes from the need for supplying a formal framework in which the 'openness' and
'incompleteness' of scientific practice and knowledge can be accommodated in a
unified way (see da Costa and French 2001). This is accomplished by extending, on
the one hand, the usual notion of structure, in order to model the partialness of
information we have about a certain domain (introducing then the notion of a partial
structure), and on the other hand, by generalising the Tarskian characterisation of the
concept of truth for such 'partial' contexts (advancing the corresponding concept of
quasi-truth).
The first step then, in order to introduce a partial structure, is to formulate an
appropriate notion of partial relation. When investigating a certain domain of
knowledge L1, we formulate a conceptual framework that helps us in systematising
and organising the information we obtain about L1. This domain is then tentatively
represented by a set D of objects, and is studied by the examination of the relations
holding among D's elements. However, we often face the situation in which, given a
certain relation R defined over D, we do not know whether all the objects of D (or
n-tuples thereof) are related by R. This is part and parcel of the 'incompleteness' of
our information about L1, and is formally accommodated by the concept of partial
relation. The latter can be characterised as follows:
Remark 1:
Notice that if R3 is empty, R is a usual n-place relation which can be identified
with R l •
3 This approach was first presented in Mikenberg, da Costa and Chuaqui 1986, and in da Costa
1986a. Since then it has been extended and developed in several different ways; see da Costa
and French 1989, 1990, 1993a, 1993b, 1995, and 1998, French 1997, da Costa, Bueno and
French 1998, Bueno 1997, 1999a, and 1999b, Bueno and de Souza 1996, and French and
Ladyman 1997, and 1999.
66 OTAvIO BUENO
We have now defined two of three basic notions of the partial structures
approach. In order to spell out the last, and crucial one-namely, quasi-truth -we
will need an auxiliary notion. The idea is to use, in the characterisation of quasi-
truth, the resources supplied by Tarski's definition of truth. However, since the latter
is only defined for full structures, we have to introduce an intermediary notion of
structure to link it to the former. And this is the first role of those structures that
extend a partial structure A into a full, total structure (which are called A-normal
structures). Their second role is purely model-theoretic, namely to put forward an
interpretation of a given language and, in terms of it, to characterise basic semantic
notions. 5 The question then is: how are A-normal structures to be defined?
Remark 2:
It should be noticed that, given a partial structure A, there might be too many
A.;.normal structures. We need to provide constraints to restrict the acceptable
extensions of A. In order to do that, we need first to formulate a further auxiliary
notion (see Mikenberg, da Costa and Chuaqui 1986). A pragmatic structure is a
partial structure to which a third component has been added: a set of accepted
sentences P, which represents the accepted information about the structure's
domain. (Depending on the interpretation of science which is adopted, different
kinds of sentences are to be introduced in P: realists will typically include laws
and theories, whereas empiricists will tend to add mainly certain laws and
observational statements about the domain in question.) A pragmatic structure is
then a triple A = (D, Ri, P) iE[, where D is a non-empty set, (R;)iEJ is a family of
4 It should be pointed out that the fact that partial relations and partial structures are partial is due to the
'incompleteness' of our knowledge about the domain under investigation-with further information
about this domain, a partial relation may become total. Thus, the partial ness modelled by the partial
structures approach is not understood as an intrinsic, ontological 'partialness' in the world-an aspect
about which an empiricist will be glad to remain agnostic. We are concerned here with an 'epistemic',
not an 'ontological' partialness.
5 For a different formulation of quasi-truth, independent of the notion of an A-normal structure and in
terms of quasi-satisfaction, see Bueno and de Souza 1996.
MATHEMATICAL CHANGE AND INCONSISTENCY 67
partial relations defined over D, and P is a set of accepted sentences. The idea, as
we shall see, is that P introduces constraints on the ways that a partial structure
can be extended.
Our problem now is, given a pragmatic structure A, what are the necessary and
sufficient conditions for the existence of A-normal structures? We can now spell out
one of these conditions (see Mikenberg, da Costa and Chuaqui 1986). Let A =
(D, Ri , P) iEl be a pragmatic structure. For each partial relation Ri , we construct a set
Mi of atomic sentences and negations of atomic sentences, such that the former
correspond to the n-tuples which satisfy Ri , and the latter to those n-tuples which do
not satisfy Ri. Let M be U ie/Mi. Therefore, a pragmatic structure A admits an
A-normal structure if, and only if, the set MuP is consistent.
As this condition makes it clear, the notion of consistency plays a crucial role in
the partial structures approach. In fact, as we shall now see, the very concept of
quasi-truth, since it depends on the existence of A-normal structures, supposes the
consistency of M and P. Having said that, let us see how this concept can be
formulated.
Definition 4. Quasi-truth:
A sentence a is quasi-true in A according to B if (i) A = (D, Ri, P)ie/ is a
pragmatic structure, (ii) B = (D', R;)ie/ is an A-normal structure, and (iii) a is
true in B (in the Tarskian sense). If a is not quasi-true in A according to B, we
say that a is quasi-false (in A according to B). Moreover, we say that a sentence
a is quasi-true if there is a pragmatic structure A and a corresponding A-normal
structure B such that a is true in B (according to Tarski's account). Otherwise, a
is quasi-false.
The idea, intuitively speaking, is that a quasi-true sentence a does not necessarily
describe, in an appropriate way, the whole domain to which it refers, but only an
aspect of it-the one modelled by the relevant partial structure A. After all, there are
several different ways in which A can be extended to a full structure, and in some of
these extensions a may not be true. As a result, the notion of quasi-truth is strictly
weaker than truth: although every true sentence is (trivially) quasi-true, a quasi-true
sentence is not necessarily true (since it may be false in certain extensions of A).
This is an important feature of this notion.
It is time now to return to the issue of mathematical change. As we shall see, it
can be addressed in a new way if we explore the resources provided by the above
framework.
The pattern of mathematical change to be provided in this section has two main
features. The first is to advance a clear aim of mathematics, in terms of which
mathematical change can be evaluated. The second is the possibility of
accommodating inconsistencies in mathematical theories, both at the theoretical and
68 OTAvIO BUENO
the metatheoretical levels, that is, not only involving particular theories but also
formal representations of mathematical change.
r will suggest here a pattern of mathematical development, whose inspiration
derives from Kitcher's (and, to some extent, Lakatos's) work, but which is
articulated in terms of the partial structures framework. The idea is that those
episodes in the development of mathematics that involve inconsistencies, ambiguous
notions, or overlap of different concepts, can be straightforwardly accommodated in
terms of partial structures. As we shall see, the information about these concepts was
only partially grasped by the structures in question. Moreover, the resulting
dynamics is articulated, or reconstructed, in terms of the search for (increasingly
more) quasi-true theories-the latter are not, of course, necessarily true. (I shall
define the notion of degree of quasi-truth in a moment.)
But first how is the quasi-truth of a mathematical theory to be evaluated? Given
the definition of quasi-truth, this is always a contextual matter: we shall determine it
in terms of a given partial structure and a given set of accepted sentences (the two
main components of a pragmatic structure or, as we may call it, since it incorporates
crucial information about a given domain of inquiry, a partial mathematical
practice). As we saw, we say that a mathematical theory T is quasi-true (in a partial
structure A) if there is an A-normal structure B in which T is true. Now is this, in any
sense, an appropriate aim of mathematics? r think it is. First, although this may not
seem to be the case at first sight, quasi-truth is not too weak. Notice that for a theory
to be quasi-true is not equivalent for it to be consistent: what is required is a model
of the extended partial relations of a given partial structure and the set P of accepted
sentences. So, pragmatic structures playa crucial role in the evaluation of quasi-
truth, and to this extent, the latter is not reducible to sheer consistency. Therefore, I
am not countenancing consistency as an aim of mathematics. This would be
inadequate on two grounds: (i) consistency is certainly too weak to accommodate
the several uses of mathematics to be taken into account by an interpretation of
mathematics. For instance, in order to explain the applicability of mathematics to
non-mathematical domains, we need more than the mere consistency of mathematics
(although we certainly need less than truth). What is required, at least according to
Field, is the consistency of the relevant mathematical theory with all internally
consistent non-mathematical theories; in other words, we need the conservativeness
of mathematics, which is more than its consistency (see Field 1980 and 1989).6
Moreover, (ii) it is question-begging to assume consistency as an aim of an activity
that is so prone to entanglement with inconsistencies. And there are good arguments
to the effect that inconsistencies may have an epistemic role not only in the
understanding of mathematics, but also in its development (the well-known
examples of the calculus and naive set theory are enough to establish this point; see
da Costa 1982, 1986b, and 1989, da Costa, Bcziau and Bueno 1998, Priest 1987, and
Mortensen 1995). This provides a second reason for the appropriateness of quasi-
truth as an aim of mathematics. Given that the underlying logic of quasi-truth is
paraconsistent (see da Costa, Bueno and French 1998), the resulting conceptual
6 For a reformulation ofField's approach in terms of partial structures and quasi-truth, see Bueno 1999a.
MA THEMA TICAL CHANGE AND INCONSISTENCY 69
7 I am using and blending together here two distinct methodological frameworks: one derived from
Laudan 1984, the other from Chiappin 1989. lowe to Laudan the idea of organising scientific debates
in terms of the interplay between aims, methods and theories. To Chiappin I am indebted for the
refinement of aims via tasks and values, and in particular, to the distinction between methodological
and meta methodological debates. None of them, however, has applied the resulting framework to the
philosophy of mathematics.
8 For the articulation of this distinction in the context of the philosophy of science, see Chiappin 1989.
70 OTAVIa BUENO
metamathematical patterns and the use of a given logic. Finally, at the theoretical
core, we find the language component as well as the set of accepted sentences and
open questions. Thus, the interaction between Kitcher's five components is made
through these methodological categories.
In order to discuss how degrees of quasi-truth can be extended in theory change,
I shall first spell out how quasi-truth admits degrees, and in what sense the latter can
be defined for mathematical theories. Let A = (D, Ri, P) iEI and A' = (D', R;, P')iEI
be pragmatic structures (thus, D and D' are non-empty sets, (R;)iEI and (RDiEI are
families of partial relations over D and D', respectively, and P and P' are sets of
accepted sentences). A encompasses (partial) information about a given domain f1 of
inquiry. Under these conditions, we say that A' will be a partial extension of A if (1)
D £; D'; (2) Ri £; R;, for every i E I; and (3) P £; P'. (Since we are dealing here with
partial relations, the condition (2) holds if Ru £; R;i, and R2i £; R2i , for every i E I.
Of course, if R3 is empty-that is, if we are considering full relations-we have the
usual notion of £;.) If A' is a partial extension of A, it is clear that it encompasses
further information about A, in the sense that 'more' relations defined over A's
domain are known to hold. Notice, however, that the relations in A' may not be full.
So a 'degree' of partiality is involved in any partial extension. It should also be
stressed that, given the pragmatic structure A, any A-normal structure B is a limiting
case of A's partial extensions. This is the case in view of the definition of an
A-normal structure, since (i) the structures A and B have the same domain, (ii) B
contains only full relations, (iii) A and B share the same set of accepted sentences P,
and (iv) the sentences interpreted in A are true in B. The point of formulating the
notion of partial extension is to accommodate the degree of partiality involved in
extending a partial structure to a full one. In fact, since we need both full A-normal
structures and the standard Tarskian notion of truth to characterise quasi-truth, there
is still a huge gap between the information we actually have about a given domain
(which is at best partial), and the one we actually need in order to provide a
semantics for the language in question.
The idea is that, as we move from a pragmatic structure A to one of its partial
extensions A', we provide more information about A, increasing, in this sense, the
degree of quasi-truth of the theory about A. More formally, we say that a
mathematical theory T' is more quasi-true than a mathematical theory T if (i) T' is
quasi-true (in a partial structure A'), (ii) T is quasi-true (in a partial structure A), and
(iii) A' is a partial extension of A. This formulation accommodates the point that the
more information a mathematical theory is able to encompass, the more quasi-true it
is. 9 In this sense, by adopting quasi-truth as an aim of mathematics, we are able to
represent mathematical changes as processes in which more quasi-true theories are
put forward. Notice that, since we are considering here mathematical theories (as
opposed to theories about the physical world), the increase in the degree of quasi-
truth depends only on the 'structural' properties of the domain under consideration,
that is, on the relations holding among the (mathematical) objects of this domain.
9 Since, as we saw, quasi-truth is weaker than truth, from the fact that a theory is more quasi-true than
another, it does not follow that it is true. This supplies one of the reasons why quasi-truth can be taken
as an aim of mathematics appropriate for an empiricist (or an anti-realist) view (see Bueno 1999a).
MA THEMA TICAL CHANGE AND INCONSISTENCY 71
More information about the latter is often obtained by establishing new relations
about it, or by extending those that were not entirely defined.
In particular, there are two important processes in order to accomplish this quest
for more information: by providing detailed proof analyses (which often bring new
concepts), and by making definitions of given concepts (more) precise. 1O When a
mathematician is trying to establish a result (by providing a proof to it), but ends up
finding a counter-example, there are two basic strategies that she can take: (a) to
deny that the counter-example is actually such (this often involves redefining some
of the concepts used up to that moment, in order to get rid of the putative counter-
example), (b) to accept the counter-example as it is, and to examine whether it
brings problems to (i) the whole theorem, or (ii) only to the tentative proof. In the
latter case (ii), the mathematician starts an analysis of the proof, trying to spot a
false lemma, shown to be false by the counter-example, trying then to resume her
proof. In the case (i), the mathematician shall refine some of the concepts used up to
that moment, in order to come up with a new version of the theorem. I I
Thus far, the notion of degree of quasi-truth has been advanced, assuming that
we have a clear grasp of what a domain of inquiry is. But how should we
characterise this notion of domain in mathematics? One possibility is to say that
each domain is determined by the kind of structures studied in it. In real analysis, for
instance, one is concerned with real numbers structure; certain algebraic structures
are the subject matter of algebra; topology examines topological structures, and so
on. To some extent, Bourbaki's conception of mathematics as the interplay of three
basic kinds of structures (algebraic, topological and of order) is an important
instance of this view (see Bourbaki 1950 and 1968). Despite stressing the role
played by the interaction between these kinds of structures, rather than by taking
them in isolation, it is crucial for Bourbaki's view to understand the domain of
mathematical theories in terms of their structural (set-theoretical) properties. In this
picture, mathematics is conceived as the study of structures, and it leads to an 'open-
10 An illuminating discussion of these two processes can be found in Lakatos 1976. In what follows, I
shall be considering some of his views.
II Without developing the details, I wish to point out that the partial structures approach supplies fruitful
tools to consider the heuristic features involved in the process of construction of proofs, i.e. the
process of trying to establish particular conjectures. The stage in which one is unsure whether a
certain conjecture holds or not-the stage in which, according to Lakatos 1976, the mathematician
tries both to prove and to refute the conjecture in question-can be modelled very naturally in terms
of partial structures. In fact, we can think of the various possible outcomes of such an inquiry as
represented in a problem-space, a very general kind of mathematical structure that generalises the
notion of state-space employed in van Fraassen's version of the semantic approach (see van Fraassen
1970, 1980 and 1989). The inquiry itself can be represented as a trajectory in such a problem-space.
The trajectory, in its tum, is represented in terms of the relations that hold between the objects of the
problem-space. Now, if such relations are partial (in which case, we will have a partial-problem-
space), they are not defined for every n-tuple of objects of the domain. Which means that the
mathematician is still uncertain about the actual truth-value of the conjecture that depends upon such
relations. The latter may either belong to the relevant domain (in which case, let us say, the conjecture
would be true), or not belong to the domain (in which case, the conjecture would be false). Thus,
Lakatos's heuristic suggestion with regard to trying both to prove and to refute an open conjecture,
being connected to the actual 'openness' of the epistemic situation of the mathematician, seems
natural enough within such a context.
72 OTAvIOBuENO
ended' view about the development of mathematics (although Bourbaki never puts
the issue in these terms). After all, the characterisation of mathematical structures in
terms of three basic kinds of structures is, of course, at best tentative: it may well be
that, in the future, new types of structures will be formulated which cannot be
reduced to any of the extant structures, nor to the interplay between them. 12 In this
sense, the characterisation provided by Bourbaki is tentative, historically
determined, and inductive, because it is based on, and generalises from, our current
mathematical knowledge to its future evolution. These are all important features,
especially for the' openness' they allow in our representation of mathematics.
My claim is that, by taking quasi-truth as the aim of mathematics and by
countenancing partial structures as the corresponding formal setting, such
'openness' can be straightforwardly accommodated. Moreover, if we then use the
methodological categories previously discussed, we have a pattern of mathematical
change that, as opposed to the standard accounts, make room for inconsistencies as
an important feature of mathematics. More importantly, given the use of a
convenient paraconsistent logic, the resulting inconsistencies do not lead to triviality
(see, for instance, da Costa 1974). In order to illustrate some of these points, I will
sketch an application of this pattern, considering the early formulation of set theory.
Two main themes can be put together in order to spell out (an aspect of) the
development of set theory: (i) Cantor's celebrated diagonal proof, and the related
distinction between inclusion and membership; and (ii) Russell's paradox, which
emerged from Russell's own concern with this distinction, and the role played by
this paradox in the subsequent development of set theory.13 According to Kanamori
(1997), the important point is that we are dealing here mostly with mathematical
moves, and that despite the 'pronounced metaphysical attitudes and exaggerated
foundational claims that have been held on [set theory's] behalf, the 'progression of
mathematical moves' has been with set theory since its beginnings (1997, 281). We
may wonder, however, in what sense these moves can be called a progression. If we
want to assign to this progression any epistemic import, we need a normative
framework in terms of which we can evaluate judgements about mathematical
change; otherwise, we end up only describing (rather than explaining) a succession
12 Even if Bourbaki has not presented his own view in this fallibilistic tone, following Newton da Costa, I
think this is probably the most sensible and charitable way of understanding Bourbaki's programme.
(For a discussion of Bourbaki's view in the context of the philosophy of science, see da Costa and
Chuaqui 1988.)
13 These themes have been presented and articulated in great detail by Kanamori, and are actually only
part of a comprehensive perspective on the development of set theory (for a fuller picture, see
Kanamori 1996 and 1997). In particular, in his 1997 paper, Kanamori also examines Zermelo's study
of functions from the power set of a set X into X (i.e. f !O (X) -t X), Zermelo's proofs of the well-
ordering theorem, and the contribution of the latter to his own articulation of set theory. In what
follows, although I will not consider Zermelo's case, I shall heavily draw on Kanamori's account,
trying to provide a general conceptual setting for the changes it describes.
MATHEMATICAL CHANGE AND INCONSISTENCY 73
of moves that have been taken in the past. One of the points of this section is to spell
out how the framework provided above can be used to fulfil this role.
As is well known, in 1891, Cantor provided his celebrated diagonal proof,
establishing that the collection of functions from a set X into a set of two elements,
say, {O, I}, has a higher cardinality than X. Today, with set theory far more
articulated, this result is considered as proving how the power set operation leads to
higher cardinalities. However, as Kanamori correctly notices, 'it would be an
exaggeration to assert that Cantor himself used power sets' (1997, 282; see also
Lavine 1994, 245). He was, in fact, extending the 19th Century concept of function,
by countenancing arbitrary functions. At this stage, Cantor had already taken an
important step that would culminate in the concept of function we adopt today. From
his inquiries in trigonometric series during the 1870' s, and in trying to extend the
theory of these series to more general classes of functions, he became interested in
arbitrary sets of real numbers. From this point, it was natural to consider any
association from the real numbers to the real numbers as a function, no matter how
arbitrary it was (see Lavine 1994, 3). This would then pave the way for Cantor to
construct the diagonalising function. He could consider all functions with a given
domain L and range {O, I}, and take these functions to be enumerated by a super-
function ~(x, z), with z being the enumerating variable. The diagonal ising function is
then g(x) = 1 - ~(x, x). We have already here a pattern of abstraction that would be
typical of, and crucial for, the subsequent development of set theory.
However, Cantor's diagonal argument could only run in terms of sets once the
distinction between inclusion (~) and membership (E) was clearly made (for a
discussion, see Kanamori 1997, 283-284). This distinction is, of course, crucial for
Frege's BegrifJsschriJt, and it was presented-not without some blunders-by
Peano, who explicitly admonished against confusing the sign 'E' with the sign '~'
(see Peano 1889, 86).14 But it was in the hands of Russell that the distinction played
a crucial role. Of course, set theory, as an abstract study of the set formation { }
operation, could only be articulated if we had such a distinction clearly established
(Kanamori 1997,284). In September of 1900, after adopting Peano's distinct signs
for membership and inclusion, Russell extended Peano's approach to the logic of
relations, defined cardinal number, and recasted some of Cantor's work. With these
results, by the rest of the year, he worked out and concluded the final draft of The
Principles of Mathematics. However, the book itself was to be published only a few
years later (see Russell 1903). The reason for this is well-known: in May 1901, in
the course of this development, Russell realised that Cantor's diagonal proof could
be changed into what we now call Russell's paradox-spotting thus a tension
between Cantor's one-to-one correspondences and Peano's inclusion / membership
distinction (see Kanamori 1997,286).
Russell's first reaction was to claim that he had found an error in Cantor's
proof. IS But he soon realised that there was no error there, but actually a paradox. In
14 As Kanamori notices (1997, 284), after clearly distinguishing between membership and inclusion,
Peano made the former follow from the latter, thus undermining the very distinction he wanted to
stress (see Peano \889,90, formula 56).
IS In a letter to Louis Couturat (from 8th of December 1900), Russell notices: 'I have discovered an error
in Cantor, who maintains that there is no largest cardinal number. But the number of classes is the
largest number. The best of Cantor's proofs to the contrary [... ] amounts to showing that if u is a class
74 OTA VIO BUENO
the final version of The Principles of Mathematics, Russell made clear how to obtain
his paradox from Cantor's proof (Russell 1903, 364-368; see also 101-107). The
idea is to adapt Cantor's functional argument into an argument about memberships
and inclusions for classes (by correlating subclasses of a class with the relations on a
class to 0 and 1). Russell could then conclude that 'the number of classes contained
in any class exceeds the number of terms belonging to the class' (Russell 1903, 366;
for a discussion, see Kanamori 1997, 287). But what does this establish?
It is usual now to present Cantor's argument as establishing that no function
f: X ~ &D (X) is bijective. This is because the set {x EX: x €it fix)} is not in the
range of f But Cantor had first proved the positive result that, for any function
f: X ~ &D (X), there is a subset of X which is not in the range off, namely, the set
{x E X: x €it fix)} just defined. Russell's argument in The Principles establishing
the connection between Cantor's proof and Russell's paradox runs then as follows
(see Russell 1903,367, and Kanamori 1997,287-288). Let us denote by V the class
of all terms, and by U the class of all classes. Russell defines a functionf: V ~ U,
such that if x is not a class, then f(x) = {x}; and if x is a class, then fix) = x. Or, in
Russell's own words:
In the case of terms and classes, for example, if x be not a class, let us correlate it with
IX, i.e. the class whose only member is x, but if x be a class, let us correlate it with itself.
[Russell 1903, p. 367]
As is well known, in order to overcome this paradox, Russell devised his theory of
!ypes (see Russell 1908, and Russell 1903, 523-528), which multiplied the
distinction between membership and inclusion all the way up in the type hierarchy,
whose number is a, the number of classes included in u (which is 2") is larger than u. The proof
presupposes that there are classes included in u which are not individuals that is, members of u; but if
u = Class [that is, the class of all classes], that is false: every class of classes is a class'. (This
passage, whose original is in French, is quoted in Kanamori 1997, 286; for a discussion, see also
Moore 1995,231.)
16 As Kanamori acknowledges (1997, 288, note 21), the formal transition from {x: x ~fix)} to {x: x ~ x}
was stressed in Crossley 1973.
MA THEMA TICAL CHANGE AND INCONSISTENCY 75
17 For the articulation of the theory of types, see Whitehead and Russell 1910-13. A historical analysis of
the development of Russell's type theory to GOdel's constructible universe can be found in Dreben
and Kanamori 1997.
18 As Russell stresses: 'It appears that the special contradiction of Chapter X Russell's paradox] is solved
by the doctrine of types, but there is at least one closely analogous which is probably not soluble by
this doctrine. The totality of all logical objects, or of all propositions, involves, it would seem, a
fundamental logical difficulty. What the complete solution of the difficulty may be, I have not
succeeded in discovering; but as it affects the very foundations of reasoning, I earnestly commend the
study of it to the attention of all students oflogic' (Russell 1903,528).
19 This point has been challenged by Lavine (1994), who claims that naive set theory is a creation of
Russell, and it is incorrectly attributed to Cantor. But let us stick, for the sake of the argument, to the
received view about the development of set theory, in which naive theory plays a crucial role.
76 OTAVIO BUENO
Of course, if we replace the formula F(x), in (a), for x t$. x, we immediately derive
Russell's paradox. In other words, the principle of separation (a) is inconsistent. As
a result, if (a) is added to first-order logic (which is often taken as the logic of the
language of set theory), the resultant theory is trivial.
We can distinguish the various classical set theories in terms of the restrictions
they impose on (a), in order to avoid the paradox. The problem, however, is that the
resulting theory might become too weak. In order to overcome this difficulty, further
axioms have to be added, besides extensionality and separation (with appropriate
restrictions). Which axioms to introduce will depend on the particular case under
consideration. For example, in Zermelo-Fraenkel set theory (ZF), separation is
formulated thus:
where the variables are subject to the usual conditions. We can say that, in ZF, F(x)
determines the subset of the elements of the set z that have the property F (or satisfy
the formula F(x)). But in the Kelly-Morse system, separation is formulated
differently:
Finally, with the notion of stratification, in Quine's NF the scheme of separation has
the form:
where the formula F(x) is taken to be stratifiable (assuming also that the standard
conditions with regard to the variables are met).
Given that all these proposals were formulated in order to avoid a given
inconsistency, we may wonder whether this problem can be considered from a
distinct point of view. Would it be possible to accept the scheme (a) without
restrictions-but without triviality? The answer is straightforward: in order to do
that, we have to change the underlying logic, so that (a) does not lead to
trivialisation. Since the separation scheme (without considerable restrictions) leads
to contradictions, this logic has to be paraconsistent. (This was in fact one of the
motivations underlying da Costa's formulation of paraconsistent logic; see da Costa
1974, and, for further discussion, da Costa and Bueno 2001.)
It was slowly verified that there are infinitely many ways to make the classical
restrictions to the separation scheme weaker, each of them corresponding to distinct
categories of paraconsistent logics. Moreover, extremely feeble logics have been
formulated, and in terms of them it is possible to use, without triviality, the scheme
(a). Some set theories, in which the forms (b), (c) and (d) of separation are either
combined or adopted in isolation, have also been constructed.
MA THEMA TICAL CHANGE AND INCONSISTENCY 77
An important point is that several paraconsistent set theories contain the classical
ones (in the formulations of Zermelo-Fraenkel, Kelly-Morse or Quine). Thus,
paraconsistency not only goes beyond the classical domain, but it also allows the
reconstruction of classical mathematics. In this respect, paraconsistent theories
provide an extension of their classical 'counterparts'. In particular, with
paraconsistency, further possibilities can be explored: for instance, properties of
Russell's set-an object which does not exist in classical theories-can be studied in
a convenient paraconsistent set theory (see da Costa 1986b, and da Costa, Beziau
and Bueno 1998). Even if not driven by the search for truth, in this picture,
mathematical change can then be modelled as the pattern resulting from the search
for ever more quasi-true theories-including those which are ultimately inconsistent.
REFERENCES
Bourbaki, N. (1950), The Architecture of Mathematics. American Mathematical Monthly 57, 231-242.
Bourbaki, N. (1968), Theory of Sets. (Translated from the French edition.) Boston, Mass.: Addison-
Wesley.
Bueno, O. (1997), Empirical Adequacy: A Partial Structures Approach. Studies in History and
Philosophy of Science 28, 585-610.
Bueno, O. (1999a), Empiricism, Conservativeness and Quasi-Truth. Philosophy of Science 66
(Proceedings),474-485.
Bueno, O. (I 999b ), What is Structural Empiricism? Scientific Change in an Empiricist Setting.
Erkenntnis 50, 59-85.
Bueno, 0., and de Souza, E. (1996), The Concept of Quasi-Truth. Logique et Analyse 153-154,183-199.
Chiappin, J. R. N. (1989), Duhem 's Theory of Science: An Interplay Between Philosophy and History of
Science. Ph.D. thesis, University of Pittsburgh.
Crossley, J. N. (1973), A Note on Cantor's Theorem and Russell's Paradox. Australasian Journal of
Philosophy 51,70-71.
da Costa, N. C. A. (1974), On the Theory of Inconsistent Formal Systems. Notre Dame Journal of
Formal Logic 15,497-510.
da Costa, N. C. A. (1982), The philosophical import of parae on sis tent logic. The Journal of
Non-Classical Logic 1, 1-19.
da Costa, N. C. A. (I 986a), Pragmatic Probability. Erkenntnis 25,141-162.
da Costa, N. C. A. (I 986b), On Paraconsistent Set Theory. Logique et Analyse 115, 361-371.
da Costa, N. C. A. (1989), Mathematics and Paraconsistency (in Portuguese). Monograjias da Sociedade
Paranaense de Matematica 7. Curitiba: Universidade Federal do Parana.
da Costa, N. C. A., J.-Y. Beziau, and O. Bueno (1998), Elements of Paraconsistent Set Theory (in
Portuguese). Campinas: Cole91iO CLE.
da Costa, N. C. A., and O. Bueno (2001), Paraconsistency: Towards a Tentative Interpretation. Theoria
16,119-145.
da Costa, N. C. A., O. Bueno, and S. French (1998), The Logic of Pragmatic Truth. Journal of
Philosophical Logic 27, 603-620.
da Costa, N. C. A., and R. Chuaqui (1988), On Suppes' Set Theoretical Predicates. Erkenntnis 29,
95-112.
da Costa, N. C. A., and S. French (1989), Pragmatic Truth and the Logic ofinduction. British Journal for
the Philosophy ofScience 40,333-356.
da Costa, N. C. A., and S. French (1990), The Model-Theoretic Approach in the Philosophy of Science.
Philosophy of Science 57, 248-265.
da Costa, N. C. A., and S. French (I 993a), Towards an Acceptable Theory of Acceptance: Partial
Structures and the General Correspondence Principle. In Correspondence, In variance and Heuristics:
78 OTAvIOBuENO
Essays in Honour ojHeinz Post, S. French, and H. Kamminga, Dordrecht: Reidel, 1993,
pp. 137-158.
da Costa, N. C. A., and S. French (1993b), A Model Theoretic Approach to "Natural Reasoning".
International Studies in the Philosophy oJScience 7, 177-190.
da Costa, N. C. A., and S. French (1995), Partial Structures and the Logic of Azande. American
Philosophical Quarterly 32, 325-339.
da Costa, N. C. A., and S. French (2001), Partial Truth and Partial Structures: A Unitary Account oj
Models in Scientific and Natural Reasoning, unpublished book, University of Sao Paulo and
University of Leeds.
Dalla Chiara, M. L., K. Doets, D. Mundici, and 1. van Benthem (eds.) (1997), Structures and Norms in
Science. Dordrecht: Kluwer Academic Publishers.
Dreben, B., and A. Kanamori (1997), Hilbert and Set Theory. Synthese llO, 77-125.
Field, H. (1980), Science without Numbers: A DeJense oJNominalism. Princeton, N.J.: Princeton
University Press.
Field, H. (1989), Realism. Mathematics and Modality. Oxford: Basil Blackwell.
Frege, O. (1879), BegrifJsschriji, eine der arithmetischen nachgebildete Formel-sprache des rein en
Denkens. Halle: Nebert. (English translation in From Frege to Godel: A Source Book in
Mathematical Logic. 1879-1931. J. van Heijenoort (ed.), Cambridge, Mass.: Harvard University
Press, 1967, pp. 1-82.)
French, S. (1997), Partiality, Pursuit and Practice. In Structures and Norms in Science, M. L. Dalla
Chiara, K. Doets, D. Mundici, and 1. van Benthem (eds.), Dordrecht: Kluwer Academic Publishers,
1997, pp. 35-52.
French, S., and 1. Ladyman (1997), Superconductivity and Structures: Revisiting the London Approach.
Studies in History and Philosophy oj Modern Physics 28, 363-393.
French, S., and 1. Ladyman (1999), Reinflating the Semantic Approach. International Studies in the
Philosophy oj Science 13, 103-121.
French, S., and H. Kamminga (eds.) (1993), Correspondence. Invariance and Heuristics: Essays in
Honour oj Heinz Post. Dordrecht: Reidel.
Hintikka, J. (1995), (ed.), From Dedekind to GOdel. Dordrecht: Kluwer.
Kanamori, A. (1996), The Mathematical Development of Set Theory from Cantor to Cohen. Bulletin oj
Symbolic Logic 2, 1-71.
Kanamori, A. (1997), The Mathematical Import of Zermelo's Well-Ordering Theorem. Bulletin oj
Symbolic Logic 3, 281-311.
Kitcher, P. (1984), The Nature oj Mathematical Knowledge. New York: Oxford University Press.
Kuhn, T. (1962), The Structure oJScientific Revolutions. Chicago: University of Chicago Press. (A
second, revised edition was published in 1970.)
Lakatos, l. (1976), Proofs and ReJutations: The Logic oj Mathematical Discovery. Cambridge:
Cambridge University Press. (Edited by 1. Worrall and E. Zahar.)
Lakatos, l. (1978a), The Methodology oj Scientific Research Programmes. Philosophical Papers, volume
1. Cambridge: Cambridge University Press. (Edited by J. Worrall and O. Currie.)
Lakatos, l. (1978b), Mathematics. Science and Epistemology. Philosophical Papers, volume 2.
Cambridge: Cambridge University Press. (Edited by J. Worrall and O. Currie.)
Laudan, L. (1977), Progress and Its Problems. Berkeley: University of Cali fomi a Press.
Laudan, L. (1984), Science and Values: The Aims oJScience and their Role in Scientific Debate.
Berkeley: University of California Press.
Lavine, S. (1994), Understanding the Infinite. Cambridge, Mass.: Harvard University Press.
Mikenberg,l., N. C. A. da Costa, and R. Chuaqui (1986), Pragmatic Truth and Approximation to Truth.
Journal oJSymbolic Logic 51, 201-221.
Moore, O. H. (1995), The Origins of Russell's Paradox: Russell, Couturat, and the Antinomy of Infinite
Number. In From Dedekind to Godel. J. Hintikka (ed.) Dordrecht: Kluwer, 1995, pp. 215-239.
Mortensen, C. (1995), Inconsistent Mathematics. Dordrecht: Kluwer.
Peano, O. (1889), Arithmetices principia nova methodo exposita. Turin: Bocca. (Abridged English
translation in 1. van Heijenoort (ed.), 1967, pp. 85-97. All references are to the English edition.)
Priest, O. (1987), In Contradiction. Dordrecht: Nijhoff.
Russell, B. (1903), The Principles oJMathematics. London: Routledge. (Second edition 1937.)
Russell, B. (1908), Mathematical Logic as Based on the Theory of Types. American Journal oj
Mathematics 30, 222-262. (Reprinted in From Frege to Godel.· A Source Book in Mathematical
MATHEMATICAL CHANGE AND INCONSISTENCY 79
Logic, 1879-1931. J. van Heijenoort (ed.), Cambridge, Mass.: Harvard University Press, 1967,
pp. 150-182.)
van Fraassen, B. C. (1970), On the Extension of Beth's Semantics of Physical Theories. Philosophy of
Science 37, 325-339.
van Fraassen, B. C. (1980), The Scientific Image. Oxford: Clarendon Press.
van Fraassen, B. C. (1989), Laws and Symmetry. Oxford: Clarendon Press.
van Heijenoort, J. (1967), (ed.), From Frege to COde/: A Source Book in Mathematical Logic, 1879-1931.
Cambridge, Mass.: Harvard University Press.
Whitehead, A., and B. Russell (1910-13), Principia Mathematica. (3 volumes.) Cambridge: Cambridge
University Press.
BRYSON BROWN
APPROXIMATE TRUTH
A Paraconsistent Account
1. INTRODUCTION
I think of myself as a realist, in at least two senses of that word. First, I (try to) face
things as they are, rather than ignore evidence that the world is not unfolding as I
would like it to. And second, I think that science aims and often succeeds at telling
us how things are. The first sort of realism is generally regarded as a (practical)
virtue-we are more likely to succeed if we work from true beliefs about the way
things stand, than if we start with wishful thinking.' But the second is pretty
controversial, and controversial for reasons related to the first: scientific realism, to
many, is itself a bit of wishful thinking. It ignores various limits on our ability to
know the world, on the notions of truth appropriate to our use of language, and the
form that our scientific theories take.
This paper is part of a larger project, working towards a new version of scientific
realism-one not so strong as to make me guilty of wishful thinking, nor so weak as
to make me (really) a non-realist after all. The focus of this paper is on one of the
central notions that many realists have appealed to: the very idea of approximate
truth. For reasons that will emerge, on my account approximate truth turns out to
require paraconsistent treatment.
Characterizations of realism in the literature have generally fallen into two camps.
Some give a prospective account, according to which the realist takes arriving at the
truth to be the aim of scientific theorizing, while the anti-realist attributes more
modest goals to science; others place the difference in the here and now, saying that
realists take (some of) our best current theories to be approximately true, while anti-
, I should add, though, that there are some interesting issues here. Some recent studies have shown
clinically depressed patients make more accurate predictions of what others will say about them when
asked in confidence than normal controls do. Perhaps a certain amount of wishful thinking (on some
topics, at any rate) is healthy rather than pathological.
81
J. Meheus (ed.), Inconsistency in Science, 81-103.
© 2002 Kluwer Academic Publishers.
82 BRYSON BROWN
realists do not. I find the first approach unsatisfactory because it makes the
difference purely prospective; realist and anti-realist could have the very same views
about the status of current theories. Simply having the long run aim of truth leaves
no present distinction between realists and (for example) constructive empiricists. A
characterization of scientific realism that makes realism a live issue now, rather than
a question about theories we do not have now, and may well never have, will need to
describe a reserved, pro-tem form of commitment that is still realistic. And it must
be such that, if realists are correct, we can reasonably make this commitment to
(some) present theories. Approximate truth is one way of trying to characterize what
such a commitment asserts about a theory.
But this approach to characterizing realism is also less than satisfactory: to date,
we lack an adequate account of approximate truth-one that can explain the partial
success of our current theories, that is recognizably related to the notion of fully-
fledged truth, and that can be plausibly attributed to theories with all the
imperfections the history of science suggests our current theories (very probably)
suffer from.
One obvious criticism begins with the observation that what is "approximately
true" is clearly not true. That is, we are not dealing with some kind of truth here, but
rather with some kind of falsehood. So we must ask just what is gained, for the
realist, by picking out a kind of falsehood and labeling it "approximate truth"? The
immediate response is to say that we aim to pick out claims (or sets of claims) that,
while false, nevertheless approach the truth, just as an ordinary measurement result,
though false, still approaches the truth about the length of a board. But this appeal to
a parallel with approximation in measurement is dubious.
In the case of a measurement we have a clear way of reporting the measurement
result that respects the tension between precision and accuracy by allowing a margin
of error (its size depending on the circumstances of the measurement and the
required confidence level). The result is a trade off between precision and accuracy,
which allows us to speak quite plainly about the result of the measurement being an
approximation (i.e. falling within a certain distance) to the actual, precise (?) length
of the object measured. We can also rank measurements as being better (or poorer)
depending on the size of their margins of error (assuming, for now, no systematic
error is involved in any of the methods being used).
Things are not nearly so clear with approximate truth. Approximate truth must
deal with failures of conceptual fit, not merely lack of perfect accuracy and precision
in a figure: the concept of "length" is not put in question by treating measurement
results as only approximations to the actual length of an object, but when we speak
of classical physics as approximately true, we recognize that the classical concept of
"mass"-to choose a famous example-does not quite apply to the objects of the
real world.
Another debate that plays an important role in this area arises from historical
developments in philosophy of science. The surrender of a sharp analytic/synthetic
distinction and the resulting demise of "meaning postulates" that had been meant to
preserve a clear distinction between observational reports and theoretical assertions
ApPROXIMATE TRUTH 83
have been a two-edged sword for realists. On one hand, it undennined empiricist
views that had relied on arguments like those in "The Theoretician's Dilemma"
(Hempel 1958) to defend the dispensability of all but instrumentalistic commitments
to theory. After all, if we cannot even state our observations without making use of a
theory's conceptual apparatus, how can we avoid making an implicit commitment to
the correctness of whatever theory we use to frame our observations? But on the
other hand, it raised problems for standard views of scientific progress: how can we
fairly compare competing theories if we have no neutral observations against which
to test them? Worse yet, as Feyerabend argued so forcefully, the notion that later
theories somehow include the truths that earlier theories had captured now seemed
untenable. Galileo's law of falling bodies is simply false in the Newtonian picture,
since the acceleration due to gravity is not constant, after all. And in general, even
when new theories allowed us to derive sentences that resembled those of earlier
theories, the notion that the sentences somehow had the same meaning seemed to be
indefensible; conceptual incommensurability ruled it out.
The need for a new account of observation was clear; one important candidate,
due to Wilfrid Sellars (1963, 321f.) was the pragmatic theory of observation. This
view accepts that observations are made using the conceptual resources of the
theoretical language. In outline, the theory holds that human beings have a capacity
to respond to various events or conditions in their environment with reports or
descriptions, but the conceptual resources they use to make such reports are a
product of linguistic training, not an in-built, fixed set of basic observational
concepts. On Sellar's account, as we acquire a language we learn to make
"language-entry" moves, in which (given that certain conditions hold to which we
can respond in this sort of way) we assert sentences without depending on the
previous assertion of other sentences to warrant our assertion. 2
Dudley Shapere also contributed to the new, more theoretically committed view
of observation. In Shapere 1982 he argued that what scientists mean by "direct
observation" is dependent on background knowledge, and involves a rich theoretical
reading of what it is that has been observed. For example, he argued that nuclear
processes at the core of the sun can be "directly" observed by means of neutrinos,
which proceed directly from the core to the earth, and whose flux carries
infonnation about conditions at the core.
This approach to observation in philosophy of science was a powerful force,
leading many to favour some fonn of scientific realism in the 1960's and 70's-see,
for example, Suppe 1977 (650ft). But with the publication of The Scientific Image
(van Fraassen 1980) the tide turned. Van Fraassen proposed a new fonn of
empiricism. He accepted the pragmatic account of observation, but found a new way
to express the empiricist intuition that there is a gap between meeting the constraints
2 Together with language entry rules, we also acquire language-language rules governing inferences, and
language-exit rules linking linguistic positions to actions in the world. This simple picture of the role
of language in our dealings with the world allows us both to say that our conceptual grasp of the
world is inescapably shaped by our language, and to recognize the world as nevertheless constraining
how we use language to grasp the world. And coping with these constraints imposes relations between
different languages, which render incommensurability less than an impenetrable barrier to
communication.
84 BRYSON BROWN
3 By "creatures like us" van Fraassen means, at least in principle, that the theory provides a description of
us and our sensory capacities that sets the limits on what we can and can't observe. So in principle the
limits on observation are determined in an anthropocentric way. It's worth noting that the examples
van Fraassen gives of aspects of models of theories that cannot be observed don't seem to be
anthropocentric in this way. Thus, for example, velocity with respect to absolute space is not just
something we can't observe according to Newtonian dynamics, it's something that nothing could
observe-it makes no difference whatsoever to relative motions, and similarly for the rotated quantum
mechanical system he describes. Physical theories have often included such empirical danglers, but
their status as danglers has generally not turned on any anthropocentric aspects of our limits as
observers.
ApPROXIMATE TRUTH 85
Scientific realists generally take the success of certain scientific theories to be the
principle source of support for the claim that they are approximately true; in the
famous "no-miracles" argument Hilary Putnam (l97l)-giving credit to R. Boyd-
urged that the success of science would otherwise be miraculous. Two key claims
are involved here: first, that the approximate truth of a theory explains its success, so
that the probability of success given its approximate truth is quite high. Second, that
no other explanation is available, so that the probability of success given that the
theory is not approximately true is very low. Given these two claims, a simple
application of Bayes' theorem seems to show that the approximate truth of the
theory is supported by its success:
4 A group including N. da Costa, S. French, Mikenberg, Chuaqui and O. Bueno have been pursuing a
more semantically oriented version of this account of observation, according to which observations
specify a partial structure which, if the theory is to meet the demands of the observations, must belong
to the complete structure of at least one model of a theory. See Mikenberg el al. 1986, da Costa el al.
1990, and Bueno and De Souza 1998. Somewhat confusingly, they sometimes describe this empiricist
notion of partial structures as a theory of approximate truth.
86 BRYSON BROWN
At this point I will be borrowing heavily from James Brown's recent Smoke and
Mirrors (l994V There Brown cites three aspects of the success of science that
might be used in defence of scientific realism:
(i) Science organizes and unifies a great variety of known phenomena.
(ii) Science has improved over time in respect of (i).
(iii) A statistically significant number of novel scientific predictions pans out (i.e.
science does better at novel predictions than mere guesswork).
Bas van Fraassen proposes that the practice of science involves a selective
process that leads scientists to accept those theories that do best at (i). Brown
criticizes van Fraassen's account first because it cannot explain (iii), and second
because the selectionist story does not explain (i) or (ii) either, unless rational theory
choice is definitionally linked to such success, which Brown rejects as
methodologically untenable. As far as this second point goes, I don't believe a
definitional link is required, so long as in the actual circumstances the criteria for
rational theory choice have in fact favoured theories that are most successful at
organizing and unifying the phenomena. So I accept that van Fraassen's approach is
tenable as an explanation of both (i) and (ii) above. But Brown's first point stands,
assuming that a good case can be made for (iii).
Making that case goes beyond the scope of this paper, but I will say a little in
support of (iii) before moving on. One example of the sort of novel prediction in
question is the violation of the Bell inequality. There is a vast weight of
presumption, based on widely successful prior practice, in favour of explaining non-
local correlations in terms of common causes within the past light cone. No one
would ever have predicted the inequality would fail without Bell's demonstration
that this was a consequence of quantum theory. Several points are worth noting here.
First, the effect involves a type of system (for example, photons from a theoretically
well-understood source) and measurement (polarization) for which the theory's
success in terms of (i) and (ii) is well-established. So it is (as far as a novel effect
can be) well within the range of conditions in which the theory has been successfully
applied in the past. Second, prior to the experiments of the last two decades or so,
the effect had no support aside from the fact that quantum mechanics predicted it.
As a result, the success of this prediction has strongly reinforced the already
5 Though Brown himself is dubious about the potential of approximate truth and its near relative,
verisimilitude-dubious for the rather weighty reason that attempts to give an account of these
notions have been notoriously unsuccessful-I don't think his stand is tenable for a realist. If our
present theories are not flat-out true, yet we want some present explanatory role for truth, we need a
pro-tern, imperfect stand-in for truth that will still carry the explanatory load.
ApPROXIMATE TRUTH 87
from empirical success so far to future success seems unjustified. Without truth or
approximate truth the Eeyore principle is inescapable.
The upshot of this discussion is that if
I. J. Brown is right to claim that our theories are more successful at novel
predictions than we would expect simply on the basis of their past empirical
success.
2. Realists can provide an account of approximate truth which explains both the
past success of our theories and these otherwise unaccounted for levels of
success with novel predictions.
then we really will have grounds for favouring realism over constructive
empiricism: theories will enjoy a kind of success which only the realist has a
plausible explanation of. Of course, a constructive empiricist could retreat to a kind
of meta-constructive empiricism, rejecting inference to the best explanation in
philosophy as well as in science, and endorsing realism as, not true, but simply the
most explanatorily powerful, empirically adequate account of the enterprise of
science. But that would be a considerable retreat for van Fraassen, who has retained
the central role of explanation in philosophy though rejecting its importance in
science. 6
6 Arthur Fine argues for just this response to the realists' use of inference to the best explanation to justify
realism, claiming that since anti-realists reject inference to best explanation in science, there's no
reason for them to accept it in philosophy either (1984). But as we have just observed there is a
tension here for any anti-realist who wants to step beyond Popper's radical inductive scepticism: a
constructive empiricist who accepts a theory is making an ampliative inference from success in
observed cases to success in other cases, but nothing in his understanding of the evidence he has
underwrites that inference, since his commitment to observation claims says only that these claims
result when we use the conceptual apparatus of the theory to make our observations. The realist, by
contrast, takes his observation claims to be (approximately) true, and so has some reason to expect
that the theory will succeed in new applications. Broadly put, then, I hold that the constructive
empiricist has no adequate response to Humean scepticism about induction, and must rest content
with Hume's response, that ampliative inference is unjustified but irresistible. The realists' position
adopts a stronger form of ampliative reasoning, and receives in tum a positive account of why
ampliative reasoning is successful. This can be viewed, incidentally, as a meta-application of
something initially noted by Ron Giere, to whit, that having made a realist commitment to one theory,
we may end up with stronger evidence supporting a further theory-and if consilience has any real
epistemic weight, this stronger evidence can (perhaps with the help of other similar links) in tum
provide further support for the initial realist commitment. Fine's NOA, in fact, seems to be pretty
realistic-it accepts the truth (including successful reference and properties corresponding to the
predicates of the theory) of a theory's description of the world, The "foot-stomping" realist he
criticizes seems to be asking for some kind of evidence-transcendent link between language a'nd an
external world; but the irenic realist takes empirical success of novel predictions to be evidence for
the relation she asserts between theory and world, and empirical failure to show the relation does not
hold, or holds only in a more limited range of contexts. The charge of (empty, foot stomping)
transcendence in the realist position can only be made good if we come up with a different theory
which is adequate to our observations, as empirically strong as the theory we started with, and that
lacks the "extra" models van Fraassen invokes to show that there is a gap between (even complete)
evidence and truth-this sort of tight fit between a theory's content and the observations we can make
is an understandable sort of epistemic ideal, but one which (as van Fraassen points out) our theories
rarely meet.
ApPROXIMATE TRUTH 89
3. CONTEXTS
7 I understand that we can now solve some of the mathematical problems that beset attempts to use old
quantum theory to account for the helium spectrum, so it may be that a wider context of application
for OQT is possible today.
8 See Thome 1994 for an account of how a theory can do this.
90 BRYSON BROWN
4. PARACONSISTENCY
As Ian Hacking (1983) and Nancy Cartwright (1983), inter alia, have noted, the
scientific theories we employ at a given time can be mutually inconsistent, and
sometimes even self-inconsistent. Yet these theories often enjoy considerable
success, even in regard to novel predictions. Consider Bohr's theory of the hydrogen
atom; the account it gave of the hydrogen spectrum included, as the theory
developed, a wide range of successful applications-to problems that had previously
been regarded as insolubilia. Yet the theory employed classical electrodynamics (in
the treatment of the radiation absorbed or emitted by the atoms, and also in the
interaction between the radiation and the instruments we employ to observe
absorption and emission spectra) while explicitly asserting that electrons in
stationary states, though they are clearly accelerated charges, do not radiate. Or
consider the tensions between geophysics and biogeography in the 1930' s, when
ApPROXIMATE TRUTH 91
There are two different ways to arrive at such a logic. The first (radical) approach
keeps the classical account of implication (a set r implies a sentence u iff any model
satisfYing all members of r guarantees the satisfaction of u), while altering the
semantics to allow classically inconsistent sets to be non-trivially satisfied. Among
these are the diatetheic logics of Routley and Priest et at., as well as other logics
requiring a revision of classical semantics but abjuring the radical notion that
contradictions can really be true. But we (that is, P. K. Schotch, R. E. Jennings and
some students and colleagues of theirs, including myself-the Canadian school of
paraconsistency) have pursued a more conservative approach. We propose a
different account of implication while preserving classical semantics (or, as it turns
out, any other semantics you might prefer; the technique we apply to cope with
inconsistency is very general). The application before us today does not call for a
dialetheic logic; to say that an inconsistent theory is approximately true is not to say
it is true tout court, or even that it could be true, tout court. Neither, I suggest, does it
call for a more modestly characterized alteration of classical semantics. So I will
employ the more conservative form of paraconsistency here, which blocks the
inferential trivialization of (some) inconsistent sets without holding that they can be
satisfied. But to make the motivation for this approach to paraconsistency plainer,
we will begin with some reflections on the classical consequence relation.
The role of preservation in logic is a familiar one: we demand of both semantic and
syntactic consequence relations that they preserve certain desirable features of
92 BRYSON BROWN
Closing f under either "1=" or "1--" produces extremely conservative extensions off.
We can say that the extended set begs no questions f leaves open, in the sense that
no consistent or satisfiable extension of f is an inconsistentlunsatisfiable extension
of the closure off under "1--" or "1=". With this point in mind, we repackage A and
B in slightly different form:
Pres: Widen the set of acceptable extensions by looking for some other
desirable) feature off that we can preserve.
This is the leading theme ofJaute de mieux reasoning. When there's no consistency
or satisfiability left to preserve, is there anything else we'd still like to preserve (at
least about some premise sets)? When consistency and satisfiability are lost, we
must find something else to preserve, or give up on reasoning altogether. 9
9 Note that this applies even to those who have extended the classical notions of consistency and
satisfiability. When and if their own, extended notions of consistency and satisfiability no longer
ApPROXIMATE TRUTH 93
The obvious property to preserve is non-triviality. But this alone doesn't shed
much light; there are too many ways to go about ensuring this, some interesting and
some uninteresting. We will look for something more specific, that will in turn
ensure preservation of non-triviality for at least some inconsistentlunsatisfiable sets.
i(f) ~ min S I Con(f, S), if this limit exists (It does not exist if f contains a
contradiction. )
~ 00, otherwise
{p, -'p} [If (p /\ -'p), since i( {p, -,p} ) = 2, while i( {p, -'p, (p /\ -'p)}) = 00
apply to a set, that set will trivialize on their logic. But the second approach allows us to reason from
any set that has a property we wish to preserve, regardless of whether it is consistent or satisfiable or
not-on the most generous extension of these notions you find defensible. The only way to avoid any
interest in the second option is to maintain that any set which has any properties at all that are worth
aiming at preserving in reasoning will tum out to be consistent/satisfiable given the right
understanding of these terms. This seems at best an odd position to take.
10 As Peter Schotch has pointed out (1993) it's reasonable to regard a tautologous set as requiring less
dividing up than a merely consistent one, since a tautologous set can be freely combined with any
consistent set salva degree of incoherence, while a consistent set cannot.
II First, do no harm. See Schotch 1993.
94 BRYSON BROWN
is both sound and completely captures the aggregation that follows from [I-. To see
how this rule works, consider the set r = {p, 'p, q}. i(f) = 2, so our rule of
aggregation becomes 2/(2 + 1), i.e. 2/3. That means we can infer from any 3
elements ofr the disjunction of pairwise conjunctions amongst them. For example,
o I- A=:>0 [I- A
{A} I-B=:> {A} [I-B
(Of course, since [I- is strictly weaker than 1-, the converse of these also holds.) The
important point I want to make before moving on is that there is nothing we have
said here about inconsistent sets that a strict classical logician should object to. The
point is one she can swallow with little or no discomfort.12 Classical aggregation is
not degree-preserving when the premise set has a degree greater than 1 and less than
00, and our new inference relation rejects it when it fails this test, replacing it with a
weaker form of aggregation that does pass the test. The classicist might, of course,
claim that forcing (i.e. [I-) is not really a consequence relation, or at least not an
interesting one. Nevertheless forcing allows us to state a closure condition that
generalizes the classical one and does not trivialize all inconsistent sets, although
any set containing a contradiction still does "explode". Furthermore, as we shall see,
this closure condition allows us to make explicit formal room for the aspect of
approximate truth that we have just been discussing: its restriction to contexts,
reflecting the patchwork nature of our pro-tern scientific grasp of the world.
Before we go on, though, I want to quickly develop a further application of this
logical machinery, which extends it to deal with (high) improbabilities, in just the
way we have already dealt with inconsistencies.
Schotch and Jenning's logic offers a systematic approach to aggregation that copes
with inconsistency much more constructively than classical logic can. Aggregation
12 A classicist will understand that what follows is true; what may cause some discomfort is the idea that
there is any interest in a turnstile that aims at preserving degree instead of consistency.
ApPROXIMATE TRUTH 95
13 This point also makes the account of reductio I offer in my 1992 more attractive than Kyburg realizes;
the increase in degree involved in a reductio applied to some new sentence together with the lottery
paradox set is from 2 to 3 or higher, not from 106 + I to 106 + 2 or higher, as Kyburg suggests in
"Adjunction" footnote, 120.
14 As before, it will not if r contains an individual sentence with probability less than or equal to 11.
96 BRYSON BROWN
As usual, r p-forces u (written r [I-p u) iff every such "covering set" for r indexed
to ~ contains an element Uj such that Uj I- u. The upshot is that no sentence that
degrades probability below the cutoff level 11 will follow. The aggregation that
results, as always, is fully characterized by closure under 2/ ip + I-the disjunction
of all pairwise conjunctions amongst any ip + 1 members of r.
Given a large series of measurements, such that each has probability greater than
1-(; of being correct and the conjunction of all has probability less than 11, we will
accept the sentences reporting each individual measurement. But the probabilistic
degree of the resulting set will be greater than or equal to 2, preventing aggregation
any stronger than we get from 2/3. Thus, closing the set of accepted sentences under
such a consequence relation allows us to accept a large set of independent
measurements without being willy-nilly forced to accept their conjunction.
5. AGGREGATION IN SCIENCE
General rules of aggregation are not all there is to aggregation-no more than
general rules of inference are all there is to epistemology. While this logic retains a
role for general rules of aggregation, we also recognize the importance of pragmatic
practice, which permits many aggregations that are not required by the general rule.
In general many individual aggregations will be (minimally) acceptable, in the
sense that adding them to the set will not increase the set's level of incoherence, but
not required, i.e. not forced by the set. Adding them to the set is an ampliative move
entirely analogous to adding any other claim not implied by the set, and must be
evaluated on the same sorts of grounds. This allows for quite a bit of aggregation,
when the joint application of the claims in question is essential to our purposes.
So aggregation in practice need not be limited by the weakness of 2 / i + 1 , nor
must it be extended to the classical rule of complete aggregation. Something in
between, driven logically by the level of our commitments' incoherence, and
practically by our need to combine some sets of sentences to get important work
done, is what is required, and what Schotch and Jennings' logic allows.
At this point a second influential argument for realism is worth raising. H. Putnam
(1973) famously observed that A-intro, though truth-preserving, clearly doesn't
preserve any of the empirical substitutes for truth. Thus, insofar as scientists endorse
reasoning using A-intro, they demonstrate a realistic commitment to their theories.
Van Fraassen's response to this argument is to argue that aggregation in science is,
as we have already observed, much more diffident and delicate than A-intro (van
Fraassen 1980, 83-87). However, this is not enough to completely disarm this
argument for realism. The key issue is whether van Fraassen can account for the
extent to which we do demand aggregation. Reflecting on this issue, he says:
ApPROXIMATE TRUTH 97
.. .it seems to me that the idea of a science consisting of a family of such theories is
really not feasible, except in the philosophically innocuous sense in which it actually
does. Suppose, for example, that we try to have a mechanics and also a theory of
electromagnetism, but no theory in which the two sorts of phenomena are both
described. Where shall we find a home then for the phenomena involving moving
charges? Electromagnetism would have to be electrostatics only. [van Fraassen 1980,
86]
But there are two ways to take this. One is purely heuristic-as we extend our
electrostatics and our mechanics, we might do so by trying to combine our
electrostatic theory with our mechanical theory, in hopes of capturing phenomena
like moving (and accelerating) charges. But for the constructive empiricist this can
be no more than an heuristic strategy-one whose success, if frequent, would be
surprising. The second is suggested by van Fraassen's straightforward reference to
"phenomena involving moving charges". This remark has an air of realism about it.
After all, to describe the phenomena that way suggests we are committed to
categorizing phenomena in terms that our two theories provide even outside the
contexts where those theories have been developed and tested. That is, we are
committed to the categories that these theories (or their improved successors)
provide having some sort of extension outside their present contexts of application.
This is a substantial commitment to unity-too much, I suggest, for constructive
empiricism to explain.
The impetus towards unity, towards the ideal of a system of claims that can be
freely and fully aggregated in the classical way, is something that constructive
empiricism provides no motive beyond convenience for. A contextually constrained
patchwork could do all that empiricism asks of science. So the goal of unity, even if
it's imperfectly achieved now (and the notion that it could be achieved or imposed
from the start seems absurd), is a further point in favour of a modest form of realism.
The issue is whether "pursuit of empirical adequacy" is an adequate goal for science,
and though van Fraassen rightly argues that conjunction-moves often don't preserve
acceptability in science, the practice of science still employs such moves too
naturally and too successfully for a strictly empiricist account to capture.
When scientists encountered black-body radiation, for instance, they attempted
to account for it by jointly applying the theories they used to account for the various
phenomena they recognized to be involved. The failure to succeed in this enterprise
also led them to modify and even withdraw their commitment to what seemed to be
adequate theories in their areas of application, something which makes sense only if
that commitment is a realistic one. Just as A-intro makes good sense for a realist and
little sense---except in a weak heuristic way-for the anti-realist, the inference from
,(A A B) to ,A or ,B also has a realist flavour. That (A A B) is not empirically
adequate in no way entails that either A or B is not empirically adequate.
It's also worth noting that we are (from the anti-realist's perspective)
surprisingly successful in producing such unifying theories, often by methods that at
least begin with aggregative moves. Bohr's approach to the hydrogen atom arose
from the combined application of a mechanical approach to the atom at equilibrium
based on Rutherford's model and a quantum picture of the interaction between the
atom and radiation. Bohr had to impose some constraints on conjunction to avoid the
98 BRYSON BROWN
obvious objection to Rutherford's model. But the result was a new and successful
theory, a theory that organized and developed phenomena and theoretical ideas that
in tum contributed to the emergence of quantum mechanics.
The key point of this logical discussion is that the account of approximate truth I
will propose here copes with both inconsistency and improbability by rejecting
classical aggregation, while preserving a form of aggregation that allows for, and
makes clear the logical costs of, the contextual limits of approximate truth. This
account admits that our rules for applying predicates and for linking predicates
together, etc., may lead to out and out inconsistencies-as when observation plus
theory predicts an observation which conflicts with the results of actual observation,
and to improbabilities-as when we try to argue from the highly probable
correctness of each member of a large set of independent measurements to the
highly improbable correctness of them all. Nevertheless these theories are effective
predictive tools because they manage, within these contextual limits, to connect
accurately with what there is and how it works.
6. APPROXIMATE TRUTH
Before I can present the account of approximate truth that I've promised here, we'll
need to consider, briefly, two aspects of the relation of a theory to the world. First,
reference is a notion which, it is widely accepted, allows for success even in cases
where the object referred to is misdescribed. So Lorentz' use of the word "electron",
at least in some contexts, could succeed in referring to electrons despite the fact that
he thought of them as classical particles. We might even consider Priestly-again, in
some contexts-to have successfully referred to oxygen or oxygen-enriched air
using the phrase "dephlogisticated air".
The restriction to contexts here is important. In more theoretical contexts, where
heavy weight was put on classical assumptions about the nature of electrons and the
connection to empirical phenomena involving electrons was remote, we might doubt
whether Lorentz was successfully referring at all. Similarly, the suggestion that
Priestly was referring to oxygen is much more plausible when he is describing (or
engaged in) procedures for producing "dephlogisticated air" and less plausible when
he is discussing his theory of combustion.
This point is also crucial to my response to Laudan (1981), who has made much
of the occurrence of "non-referring" terms in successful scientific theories. Consider
the electromagnetic ether, for example. In the context of the various mechanical
models of the ether that were explored in the late 19th century, it is quite natural to
say that there simply is no such thing. But in the context of discussions of Maxwell's
equations and their actual applications, we might hold instead that the "ether" was
simply space itself, or the vacuum-after all, it is points in space-time that are
assigned the field characteristics Maxwell's equations and their quantum-mechanical
successors require, and it is an oft-made observation that the vacuum in
ApPROXIMATE TRUTH 99
Second, we also have to deal with predicates and relations and their use in
describing the things we refer to. This is much tougher going; the relation between
predicates and the world has been a bone of contention from the very beginnings of
philosophy. Recently some realists, including M. Tooley (1988), D. Armstrong
(1983), F. Dretske (1977) and 1. Brown (1994), have argued for an account invoking
Platonist universals corresponding to the predicates of science, and bearing second
order contingent relations to each other that are the metaphysical basis of natural
laws. However, it's not clear to me that this program provides enough advantages to
justify its ontological cost. Further, I'm troubled by the notion that the universals
appearing in scientific theories can be identified independently of the laws they
participate in, which is what makes the relations the laws impose on them
contingent. Finally, without some causal relation between us and these universals,
it's hard to see how we come by our "grasp" of them, and how that grasp in tum
constrains our thought and actions.
I will pursue a less metaphysically committed approach in this account of
approximate truth. I assume only that established methods for applying predicates
often succeed in linking them to a real property of the items we refer to, a property
which corresponds-in those contexts-to our use of the predicate. The
measurement contexts in which we employ these predicates are often governed by
formalized procedures for obtaining high-quality, replicatable results. When we do
succeed in this way, the items we apply the predicate to very probably have the
property, and (where a numerical measure is involved) the result (again, with high
probability) falls within a numerical approximation of the value of the property the
item actually possess (allowing for scales and similar arbitrary differences). 15
Following Sellars, I regard our use of predicates as the result of a context-of-
application constrained projection from properties and relations of items in the
world to assertions that the items have the properties. These assertions are made by
15 I don't believe this mode of speech commits me to Platonism, although a more modest realism about
universals seems inescapable-otherwise the claim just made will be trivial. This modest realism
merely acknowledges that the divisions we mark in the world with our predicates do not constitute
arbitrary collections, that in some sense (as Zenith used to have it in their famous ad slogan) the
quality goes in before the name goes on, and that a pure nominalism, rejecting anything beyond
arbitrarily specifiable extensions as the semantic content of predicates and relations, is not tenable.
See Wilfrid Sellars on universals and distributive singular terms (for example, Sellars 1963, Ch. 8),
where he argues that predicates are, first, words that play roles in our language, which roles can
parallel those of words in other languages, justifying the use of a distributive singular term correctly
applied to all words in all languages playing similar roles. The constraints on what predicates are
natural and what predicates are "gerry-mandered", etc. (see also Lewis 1986) derive from our ability
to learn to use words that play the role these do-which involves both anthropocentric facts about us
(colour terms come to mind here) and facts about how the world work (including our abilities to
respond differentially to features of the world) that makes "carving nature at the joints" an end we can
(at least in part) hope to achieve.
100 BRYSON BROWN
juxtaposing predicates and referring terms; what they then assert is that the terms'
referent is characterized in a way corresponding (under the projection rules) to the
character we give the term by juxtaposing it with the predicate (Sellars 1963, and
1979).16
It's important to recognize here that what has to be "out there" in correspondence
to what the theory asserts is quite schematic. Even if we are talking about full-
fledged truth, all we need (as van Fraassen says) is that the world be a model of the
theory, i.e. that there be items which we can take the theory's terms to refer to, and
properties and relations that the items possess which can serve as interpretations of
the predicates, such that under this interpretation the sentences of the theory are
satisfied. 17 And for approximate truth, we will need the same, with the proviso that
these connections of reference and properties will be context-dependent.
For simple bivalent predicates I assume that approximate truth requires, in the
contexts of application, a high level of agreement between the application of the
predicate and the presence of the property; including a role for numerical
approximation in numerically measured properties like length, mass, etc.
Connections between these attributions and their theoretical consequences depend
on not having to divide contexts of commitment too finely. For example, the
relations classical dynamics imposes between mass, force and acceleration make the
relation between a classical physicists' use of "mass" and the corresponding
property clearly context dependent. When we step outside the range of relative
velocities, densities, or precision at which these relations produce reliable results we
are aggregating our commitments too strongly. But when we arrive at a rich body of
successful applications including observation-based predicate attributions,
theoretical inferences and successful predictions, then the division of contexts
required to make our assertions approximately true is coarse enough to allow
considerable levels of local aggregation. This is the point at which it becomes
reasonable to speak seriously of the theory as approximately true. But what we mean
by this is always something context dependent, and not a fully aggregative
commitment that unites all the theory's content in all contexts.
16 See "Truth and 'Correspondence"', Chapter 6 in Sellars 1963, esp. 211 ff.; see also Sellars 1979.
17 I do not follow Putnam's recent views, which hold that there are no further constraints on which
interpretations are admissible, and neither do I share van Fraassen's sense that we can, in these
contexts, ignore the possibility of unintended interpretations. But I will leave aside a serious
discussion of interpretation here, and simply assume that the actual application of a theory does in fact
impose constraints other than satisfaction on what can be an acceptable interpretation.
ApPROXIMATE TRUTH 101
Approximate Truth 1:
111- Pal ... an iff (Vc( al), ... , Vc( an» E Vc(P)
Where Vc( a) is the item referred to by "a" in context C, and Vc( P) is the extension
of the property attributed to a by the use of "P" in context C. To extend this account
to deal with numerical approximations, where Vc( P) expresses a quantity, we must
specify the required approximation along with the context, roughly as follows:
Approximate Truth 2:
IIr Pal ... an iff ( VC(al), ... , Vc(a n» E Vc(P) ± A
Finally, when the context of measurement only makes the approximate truth of an
assertion probable, we may also specify the level of probability involved:
For each of these the logic of "lIr" is weakly aggregative: contradictions cannot be
approximately true, and taking a contradiction as a premise leads to trivialization.
But many incompatible theories could all be approximately true-and even a single,
self-inconsistent theory could be approximately true, so long as its inconsistencies
are isolated from each other in separate contexts of application, as in Bohr's theory
of the hydrogen atom. In all these cases, the weakening of aggregation blocks
trivialization. '8
With this in hand (including a minimum probability threshold in cases where we
want to avoid improbable conjunctions as well as inconsistent ones) we now have a
notion of approximate truth. It is quite liberal, in that it recognizes some degree of
approximate truth for any theory we can apply in a regular way so that some of its
assertions correspond reliably with the presence of items with certain properties or
relations. It accepts that even very limited degrees of success indicate that this has
been achieved, in some small way. But as scientists uncover a theory's limitations,
they find that what works with some reliability in a narrowly circumscribed set of
contexts cannot be successfully extended to a less context-restricted account. We
can then move to contain the damage, applying our logic to limit the aggregation of
18 There are links here to modal logic as well-see Schotch and Jennings 1980. But we wi11leave these
aside [or now.
102 BRYSON BROWN
REFERENCES
Apostoli, P. and B. Brown (1995), A Solution to the Completeness Problem for Weakly Aggregative
Modal Logic. Journal of Symbolic Logic 60,832-842.
Armstrong, D. (1983), What is a Law ofNature? Cambridge: Cambridge University Press.
Brown, B. (1992), Rational Inconsistency and Reasoning. Informal Logic XIV, 5-10.
Brown, B. (1993), Old Quantum Theory: A Paraconsistent Approach. PSA 1992, Vol. 2, Lansing:
Philosophy of Science Association, pp. 397-411
Brown, J. (1994), Smoke and Mirrors. London: Routledge and Kegan Paul.
Bueno, O. and E. De Souza (1996), The Concept of Quasi-Truth. Logique et Analyse 153-154,183-199.
Cartwright, N. (1983), How the Laws of Physics Lie. Oxford: Oxford University Press.
Da Costa, N. e. A. and S. French (1990), The Model-Theoretic Approach to Philosophy of Science.
Philosophy of Science 57,248-265.
Dretske, F. (1977), Laws of Nature. Philosophy of Science 44,248-68.
Fine, A. (1984), The Natural Ontological Attitude. In Scientific Realism, J. Leplin (ed.), Berkely:
University of California Press.
Hacking, l. (1983), Representing and Intervening. Cambridge: Cambridge University Press.
Hempel, C. G. (1958), The Theoretician's Dilemma. In Minnesota Studies in Philosophy of Science, H.
Feigl, M. Scriven, and G. Maxwell (eds.), Vol. II,. Minneapolis: University of Minnesota.
Republished in Aspects of Scientific Explanation, e. G. Hempel, New York: The Free Press, e.G.,
1965, pp. \73-222.
Kyburg, H. (1997), The Rule of Adjunction and Reasonable Inference. Journal of Philosophy XCIV,
109-125.
Laudan, L. (1981), A Confutation of Convergent Realism. Philosophy of Science 48, 19-48.
Lewis, D. (1986), The Plurality of Worlds. Oxford: Blackwell.
MacIntosh, D. (1994), Partial Convergence and Approximate Truth. British Journal of Philosophy of
Science 45, 153-70.
ApPROXIMATE TRUTH 103
Mikenberg, 1., N. C. A. Da Costa, and R. Chuaqui (1986), Pragmatic Truth and Approximation to Truth.
The Journal of Symbolic Logic 51,201-219.
Putnam, H. (1971), Philosophy of Logic. New York: Harper and Row.
Putnam, H. (1973), Explanation and Reference. In Conceptual Change, G. Pearce, and P. Maynard,
(eds.), Dordrecht: Reidel, pp. 199-221.
Schotch, P. K. (1993), Paraconsistent Logic: The View from the Right. PSA 1992, Vol. 2, East Lansing,
Philosophy of Science Association.
Schotch, P. K. and R. E. Jennings (1980), Inference and Necessity. Journal of Philosophical Logic 9,
327-340.
Schotch, P. K. and R. E. Jennings (1989), On Detonating. In Paraconsistent Logic: Essays on the
Inconsistent, G. Priest, R. Routley, and J. Norman (eds.), Munich: Philosophia Verlag, 1989,
pp.306-327.
Sellars, W. (1963), Science, Perception and Reality. London: Routledge and Kegan Paul.
Sellars, W. (1979), Naturalism and Ontology. Reseda California: Ridgeview Pub. Co.
Shapere, D. (1982), The Concept of Observation in Science and Philosophy. Philosophy of Science 49,
485-525.
Suppe, F. (1977), The Structure of Scientific Theories. Second Edition, Chicago: University of Illinois
Press.
Thorne, K. S. (1994), Black Holes and Time Warps. New York: W. W. Norton & Co.
Tooley, M. (1988), Causation: A Realist Approach. Oxford: Oxford University Press.
van Fraassen, B. C. (1980), The Scientific Image. Oxford: Oxford University Press.
Weston, T. (1992), Approximate Truth and Scientific Realism. Philosophy of Science 59, 53-74.
NEWTON DA COSTA AND STEVEN FRENCH
* This is a condensed version of a chapter of our (hopefully!) forthcoming book on partial truth and
scientific practice.
1 See, for example, Gotesky's clear and comprehensive review of 'unavoidable' inconsistencies in moral,
'theoretical', 'factual', psychological and 'existential' situations (Gotesky 1968).
105
1. Meheus (ed.), Inconsistency in Science, 105-118.
© 2002 Kluwer Academic Publishers.
106 NEWTON DA COSTA AND STEVEN FRENCH
The existence of inconsistency poses obvious problems for any view which
construes theories as sets of sentences expressed in terms of the classical predicate
calculus, precisely because of the explosion into triviality. The 'logic driven'
approach noted above typically retains this syntactic view of theories but abandons
the underlying classical logic for some non-classical counterpart. The obvious
candidate for such an underlying framework would be one of the (infinitely) many
systems of paraconsistent logic (Arruda 1980; da Costa and Marconi 1989). As is
well known, paraconsistent logics can be regarded either as complementary to
classical logic, to be employed when we are threatened with an explosion into
triviality, or as a kind of heterodox logic, incompatible with the classical system and
put forward in order to ultimately replace the latter.
A representative example of the latter approach is Priest, who argues for the
adoption of the Hegelian position that there are true contradictions (1987). 2 The
classical generation of all possible propositions from an inconsistent theory is then
blocked through the adoption of a form of 'dialetheic' logic in which certain
contradictions are not only tolerated but regarded as true. 3 More importantly for our
purposes, Priest includes among the pragmatic grounds for adopting his position the
occurrence of inconsistency in the history of science, explicitly citing Bohr's theory
with its inconsistent mixture of classical and quantum principles (Priest 1987, 127).
Thus, the 'logic of science', for Priest, is dialetheic. Clearly, then, for an example
from the history of science to count as support for Priest's position, it must be
regarded as not only inconsistent but also as true in the correspondence sense, since,
as we have just mentioned, this is how dialetheic logic regards contradictions.
However, it is interesting to note that at the time Bohr's theory/model was proposed
there was considerable controversy over how seriously to take it (for the spectrum of
2 cf.Arruda 1980, 4.
3 Thus dia1etheism is stronger than paraconsistency in general. For a discussion of the differences
between da Costa's and Priest's approaches to such logics, see da Costa and French 1989b.
INCONSISTENCY IN SCIENCE: A PARTIAL PERSPECTIVE lO7
reactions see Pais 1991, 152-155).4 Bohr himself is recorded as saying 'No you can't
believe that. This is a rough approach. It has too much of approximation in it and it
is philosophically not right' (cited in Pais 1991, 155).5 Assertions of truth in this
case seem particularly inapt.
Brown has also recently espoused the use of a form of paraconsistent logic in
this context, while rejecting the position that inconsistent theories as a whole should
be regarded as true in the correspondence sense (Brown 1990). At the heart of this
view is a 'contextual' characterisation of acceptance, in the sense that a theory is
regarded as accepted when,
... we choose to treat [it] as if true, observing contextual limits on that commitment so as
to avoid bringing incompatible claims into play at any point, ... [Brown 1990, 285]
The idea here is that the context of application of an inconsistent theory may be
broken down into subcontexts in which the conflicting principles laying behind the
inconsistency may be isolated. In this way the conjunctive application of
contradictory claims is effectively blocked (Brown 1990, 284). Since there is no
context in which all the principles are collectively true, in the sense that the evidence
supporting scientific theories is essentially local in nature, the logic of acceptance
here is a paraconsistent 'non-adjunctive' one according to which the closure of a set
of claims under conjunction is not implied by the set (Brown 1990, 289-292).
As in Priest's case, Brown draws on the example of Bohr's theory to illustrate
this approach (Brown 1992).6 In particular he claims that Bohr was very careful to
distinguish the contexts in which his contradictory principles were to be applied.
Thus, on Bohr's model an electron in a hydrogen atom could exist in one or other of
a discrete set of orbits or 'stationary states'. Classical mechanics could then be
applied to account for the dynamical equilibrium of the electron in one of these
stationary states but not to transitions between states, where the relation between the
amount of energy and frequency of radiation emitted is given by Planck's formula
(Bohr 1972, Vol. 2, 167).
As Brown notes,
This combination of classical and non-classical principles is a logically risky game.
Whatever form of commitment to these principles Bohr was proposing, it was not a
commitment closed under the classical consequence relation ... [Brown 1992, 399]
Nevertheless, commitment there was and Brown cites cases involving the
confirmation of the theory by various pieces of evidence in which what he calls
'literal commitment' played an important role (1992, 400).7 'Literal commitment'
4 Thus Moseley wrote, ' ... I believe that when we really know what an atom is, as we must within a few
years, your theory even if wrong in detail will deserve much of the credit.' (Bohr 1972, Vol. 2,
545-546). Ehrenfest was, characteristically perhaps, much more negative, writing that 'Bohr's work ...
has driven me to despair. If this is the way to reach the goal I must give up doing physics.' (Klein
1970,278) and referring to the 'completely monstrous Bohr model ... ' (Klein 1970,286).
5 Einstein said of Bohr: 'He utters his opinions like one perpetually groping and never like one who
believes he is in possession of definite truth ' (quoted on the frontispiece of Pais 1991).
6 Pais refers to Bohr's model of the hydrogen atom as a 'Triumph over logic' (Pais 1991, 146).
7 It perhaps needs pointing out that the empirical support for Bohr's theory was extremely strong. Pais,
for example, notes that the 'capstone' of Bohr's work on the Balmer formula for the spectrum of
108 NEWTON DA COSTA AND STEVEN FRENCH
involves belief in the truth of the separate components, or 'patches', of the theory.
According to Brown, then, Bohr was clearly committed to both the quantisation of
radiation energy and classical physics. Applying a classical closure relation on this
set of inconsistent commitments, under which there is adjunction so that truth is
effectively distributed, leads to triviality. Hence Brown advocates the imposition of
a non-adjunctive logic which supplies a non-trivial form of closure:
This modest proposal regarding closure allows us to take Bohr's commitment to these
inconsistent principles seriously and at face value, without regarding him as committed,
implicitly, to anything and everything. [Brown 1992,405]
atomic hydrogen is represented by his theoretical account of the so-called 'Pickering' lines, which
Bohr ascribed to singly ionised helium. When it was objected that Bohr's account failed to fit the data,
Bohr responded that his theory was based on an approximation in which the nucleus was treated as
infinitely heavy compared to the electron and that with more realistic masses included 'exact
agreement' with the data was obtained. As Pais says, '[u]p to that time no one had ever produced
anything like it in the realm of spectroscopy, agreement between theory and experiment to five
significant figures.' (Pais 1991, 149).
INCONSISTENCY IN SCIENCE: A PARTIAL PERSPECTIVE 109
Bohr actually presented three different treatments of the hydrogen atom (see Pais
1991, 146-152) but what they all have in common is the postulate that an electron in
such an atom is restricted to one or other of a discrete set of orbits or 'stationary
states'. Now Brown emphasises the point that classical mechanics was taken to
apply to the dynamics of the electron in its stationary state, while quantum theory
was brought into play when the transition between discrete states was considered-
this discreteness contradicting classical physics, of course-and this certainly does
seem to mesh nicely with the non-adjunctive view. However, the conflict between
quantum and classical physics is also manifested in what has been called 'one of the
most audacious postulates ever seen in physics' (Pais 1991, 147), namely the
assertion that the ground state was stable, so that an electron in such a state would
not radiate energy and spiral into the nucleus as determined by classical physics.
This is the central inconsistency of the Bohr model but it is there right in the heart of
the context in which the governing principles are those of classical dynamics. 8 Of
course, in applying and developing the old quantum theory, Bohr, Ehrenfest,
Sommerfeld9 and others imported and employed claims from classical mechanics-
this was characteristic of the 'old' quantum theory and indeed this was surely the
only way to proceed, at least until the work of SchrOdinger and Heisenberg-but
Bohr's theory cannot be so easily broken down into distinct contexts or 'cells', to
each of which one can systematically assign principles held as true. to
The alternative to the 'logic driven control' of logical anarchy is the so-called
'content driven' view. According to this, the attitude of scientists to inconsistency is
based simply on ' ... a reflection of the specific content of the physical theory at
hand.' (Norton 1993, 417). Thus Norton (1987) argues that in the case of the
8 von Laue is recorded as exclaiming, 'This is all nonsense! Maxwell's equations are valid under all
circumstances. An electron in a circular orbit must emit radiation.' (Jammer 1966, 86). Einstein, on
the other hand, said, at the very same occasion (which was a joint physics colloquium of the
University and Institute of Technology in Zurich), 'Very remarkable! There must be something
behind it. I do not believe that the derivation of the absolute value of the Rydberg constant is purely
fortuitous.' (Jammer 1966).
9 Sommerfeld, of course, greatly extended Bohr's theory and indeed, the theory came to be known as the
'Bohr-Sommerfeld theory of atomic structure'. Interestingly, Sommerfeld's previous work on the
dispersion of light in gases was criticised by Oseen in a letter to Bohr for being inconsistent since it
was based on Bohr's model together with Lorentz' theory of the electron (Bohr 1972,337). In reply,
Bohr also rejected Sommerfeld's dispersion theory but on experimental and theoretical grounds (Bohr
1972).
10 Hettema seems to miss the point in writing that, ' ... arguably, the Bohr theory is not inconsistent with
the Maxwell-Lorentz theory, in that it uses a much lesser portion of the Maxwell-Lorentz theory than
Lakatos contends .. .' (1995, 323). cf. Bartelborth 1989. Hettema is correct in claiming that Lakatos
overstates his case when he asserts that Bohr's programme was 'grafted' onto Maxwell-Lorentz
electrodynamics, since, as Brown notes, Bohr separated the frequency of the emitted radiation from
the frequency of the periodic motion of the electrons in the atom, with quantum mechanics applying
to the former and classical mechanics to the latter (Hettema 1995,314). Nevertheless, the posited
stability of the ground state is in direct conflict with a fundamental component of classical
electrodynamics, namely that accelerating charges radiate.
110 NEWTON DA COSTA AND STEVEN FRENCH
like Norton, Smith draws on the example of the 'old' quantum theory; however,
whereas Norton was primarily concerned with reconstructing a consistent
formulation of this theory, Smith is quite clear that his focus is on the issue of the
grounds for believing that such formulations even exist when all that a scientist has
is the original inconsistent 'proposal' (l988b, 435, fn. 20). As Einstein clearly saw,
Planck's inconsistent derivation of the black body radiation law was inadequate but
also contained an element of truth (Smith 1988b, 435).13 This was subsequently
expressed in Einstein's famous 1917 paper on the spontaneous and stimulated
emission of radiation as the conditions for equilibrium between emission and
absorption processes that would have to be obeyed by both classical and quantum
treatments (Smith 1988b, 436-437). However, although Planck's inconsistent
treatment contained this 'element of truth', Smith prefers not to regard it as a fully
fledged theory. Relatedly he argues that we need something weaker than acceptance
to refer to the attitude adopted towards such 'proposals'.
Smith would certainly agree with Brown's requirement that the division into
contexts of sub-sets should be such as to allow one to use certain claims in those
sub-sets where they are supported by evidence. However, it might be asked whether,
without the imposition of closure, this is sufficient to prevent the division into sub-
sets from becoming an unsystematic affair, with the application of claims from these
sub-sets becoming completely unregulated. Certainly abandoning closure is a radical
move to make under any circumstances since it places these scientific developments,
profoundly important as they are, beyond the reach of logic altogether.
At this point the possibility of question begging by both sides looms. Smith
wants to show that we can accommodate inconsistency in science without
introducing an underlying paraconsistent logic; the price of such an accommodation
is the twin distinctions between theories and 'proto-theories' and acceptance of the
former and 'entertainment' of the latter. Our contention is that we can avoid paying
such a price by regarding theories, models, whatever, as partially true only and tying
our notion of acceptance to that (French 1991; da Costa and French 1993a).
that an objective of economics and the social sciences, but crucially not of science itself, might be to
find 'maximally inconsistent'. factually adequate 'arguments' (Madan 1983). Such an objective is
rational given the volatility of the economic environment, for example, where propositions are to be
regarded as strictly false of the actual world but 'true' enough to warrant attention. The possibility of
two theories which are minimally true in this sense yet mutually inconsistent is then analysed in terms
of a recursion theoretic formalisation of the relationship between theories and evidence. Interestingly
Madan believes that when a theory is put forward in science it is regarded as 'universally true' and
incapable of being found to be false. Given what we say here, Madan's approach might also be
profitably applied to scientific theories.
13 Smith notes that at the 1911 Solvay conference Poincare expressed his concern over the use of
contradictory principles in such derivations, pointing out the logical anarchy that (classically) results
(Smith 1988b, 431 fn. 10). Gotesky records that Poincare also said that contradiction is the 'prime
stimulus' for scientific research (Gotesky 1968,490); these two aspects of Poincare's thought are not
necessarily inconsistent of course!
112 NEWTON DA COSTA AND STEVEN FRENCH
The central notion of our account is that of a 'partial structure'. Such structures (of
first-order) can be represented in the following form:
where A is a non-empty set, and Ri, i E I, is a family of relations, which are 'partial'
because any relation Ri , iE I, of arity ni is not necessarily defined for all ni -tuples of
elements of A. By defining a 'simple pragmatic structure' in these terms, a form of
'pragmatic' or 'partial' truth can be constructed, in which the simple pragmatic
structures perform an analogous role to the set-theoretic structures in Tarski's theory
(Mikenberg, da Costa and Chuaqui 1986). This formal characterisation of 'partial'
truth has been further developed and employed by da Costa and French in a variety
of contexts (see, for example, da Costa 1989; da Costa and French 1989a, 1990,
1993a and 1993b); we will not repeat the details here.
There are a number of consequences which follow from introducing both this
notion of partial truth and, more generally, partial structures into this context. First
of all, we can regard a theory, and in particular, an inconsistent theory, taken as a
whole, as partially true and accepted as such, where this involves an appropriate
sense of commitment (see da Costa and French 1993a). Hence, contrary to Brown,
an inconsistent theory does not have to be artificially divided into consistent sub-sets
in order to capture this sense of commitment and acceptance. Furthermore, such
theories do not have to be epistemically segregated and flagged as 'proto-' theories
as Smith argues. They can be believed-as partially true-and accepted just as any
other theory. Closure is retained but not surprisingly it is non-classical: the logic of
partial truth is paraconsistent (da Costa, Bueno and French 1998) and a
paraconsistent logic of scientific acceptance is currently under development based
on this work.
Secondly, from the structural perspective, partial structures provide a means of
capturing the manner in which inconsistent theories give rise to consistent
successors. From this latter standpoint the interesting issues involve questions such
as how inconsistent theories are used for computational purposes, for example, or
how such theories might be heuristically fruitful. Unfortunately, there has been little
discussion of these issues within the semantic or model-theoretic approach. Within
this dearth of consideration Laymon stands out for declaring that one of the
advantages of this approach to theories is that it
... allows for easy and revealing consideration of computational concerns and ... allows
for the coherent use within a calculation of inconsistent theories. [Laymon 1988,
262-263]
Laymon's primary concern is with the way in which theories which are inconsistent
with one another are used to account for phenomena, rather than the explanatory
INCONSISTENCY IN SCIENCE: A PARTIAL PERSPECTIVE 113
14 Although Laymon (1985) has discussed Stokes' use of inconsistent principles in the latter's analysis of
pendulum motion.
15 We recall the enormous empirical support enjoyed by Bohr's theory for example.
16 cf. Gotesky on 'internal' and 'external' contradictions (Gotesky 1968,484).
114 NEWTON DA COSTA AND STEVEN FRENCH
Obviously we would agree with this last line but would question why an epistemic
attitude different from that of acceptance needs to be marked out for these cases,
where acceptance need not be understood as 'unconditional'. It might seem that we
merely have a difference in terminology here-where we use 'acceptance as
partially true', Smith uses 'entertainment'-but the difference is significant: as we
have already said, dropping closure puts inconsistent 'proposals' beyond the reach
of logic altogether. Thus this position rests on a dichotomy, with on the one side,
consistent theories, accepted as true and subject to classical logic, and on the other,
inconsistent proposals which are merely entertained and not closed under logical
implication.
On this view, then, inconsistent proposals do not lie within the domain of
acceptance at all but rather occupy the realm of 'discovery' (this is made clear by
Smith himself; Smith 1988a). This distinction standardly marks a separation of
discovery from justification in terms of rationality: the former ' ... can be analysed
only psychologically, not logically... and cannot be portrayed by a rational
procedure .. .'17 (Reichenbach 1949,434), whereas the opposite is true of the latter
which, it is claimed, is-or should be-the sole concern of epistemology. The
distinction itself has come under pressure from two directions: historians and
sociologists of science have questioned whether so called 'external', social factors
can be contained within 'discovery' and have argued that they spill over into
justification, collapsing any distinction between the two. Moving the other way, a
number of philosophers of science have argued that discovery is susceptible to
rational analysis and have articulated what has been called the 'theoretic' guidelines
to new theories (see, for example, Post 1971).
The heuristic role played by inconsistent theories clearly stands in need of
further elucidation. As Post, for example, has noted, an inconsistency is usually
regarded as a 'formal flaw' which points the way to the construction of a consistent
successor (Post 1971, 221-225). These flaws, or 'neuralgic points', are ' ... useful
hints of a higher and wider system in which these contradictions are eliminated'
(Post 1971,222). The process is taken to be analogous to that of moving from the
liar paradox, through Godel incompleteness to truth at the meta-level in logic. Of
17 cf. Popper 1972, 31: 'The initial stage, the act of conceiving or inventing a theory, seems to me neither
to call for logical analysis nor to be susceptible of it.'
INCONSISTENCY IN SCIENCE: A PARTIAL PERSPECTIVE 115
course a 'dialetheic realist' might question precisely such a move, arguing that truth
can be retained at the object level of the paradox itself.18 An instrumentalist, on the
other hand, while dispensing with truth altogether, would also reject the process
wholesale, arguing that inconsistent theories should not be ruled out as an ultimate
objective of science. Both views disdain the elimination of contradictions that Post
emphasises and both fall afoul of the actual practice of science where attempts to
move from inconsistency to consistency are clearly evident. 19
The general strategy that can be identified in this process is that of 'projection',
in the sense that the scientist
... uses the original [inconsistent) proposal along with the confirming evidence available
for various parts of that proposal to give a schematic 'projection' of what the consistent
replacement theory (or some fragment of that theory) will look like. [Smith 1988b, 438)
The same strategy is exemplified in Bohr's use of the correspondence principle and
provides a rational procedure for the heuristic use of inconsistency. 20
In discussing these strategies, it is significant that Post highlights the incomplete-
ness that is an essential aspect of the process, emphasising that
[t)his incompleteness of theories, or their openness to growth beyond themselves, is a
stronger and more interesting case than just filling the gaps in the data covered by a
theory, such as missing elements in the periodic table. Just as the system of real
numbers is not closed, in the sense that one can formulate equations in terms of real
numbers only whose solutions point outside the system, so a physical theory may show
its lack of closure. The points at which a theory is open, or anomalous, provide a strong
indication of the way in which the future larger theory is to be built ... [Post 1971, 222)
18 Priest claims that the application of Godel's first incompleteness theorem to the 'naive' notion of proof
constitutes a further argument for dialetheism because it shows that our naive proof procedures are
inconsistent (Priest 1987, 49-63). If we insist that naive proof theory be consistent then we must
accept a certain incompleteness, either expressive or proof-theoretic.
19 Bohr himself was greatly concerned with the logical consistency of his model. In response to Oseen's
demand for a ' ... clear logic which always remembers what the basic assumptions are, and which
never combines them with opposite assumptions for the sake of accidental progress ... ' Bohr wrote,
'As you, I feel very much indeed that a logical critic [of quantum theory) is very much needed, but I
think I take a more optimistic view on the Quantum theory than you. In fact, I have written a paper in
which I have made an attempt to show that the assumptions of the theory can be put on a logical
consistent form, which covers all the different kinds [of) applications (by this I mean all applications
which assume a discontinuous distribution of the possible (stationary) states; but I do not think that a
theory of the Planck type can be made logical consistent).' (Bohr 1972, 340 and 571). After receiving
copies of Sommerfeld's work Bohr abandoned the publication of the paper he mentions, although it
was incorporated in his 1918 account of the quantum theory of line spectra (Bohr 1972, 342 and 572).
In his letter to Sommerfeld, Bohr makes it clear that the 'logically consistent form' of which he
speaks is based on Ehrenfest's 'adiabatic principle' (Bohr 1972,604); for a consideration of the role
of this principle within the non-adjunctive framework, see Brown (1992, 401-402).
20 'The projected theory serves as a guide for the search for the material content of the proper replacement
for the original inconsistent proposal.' (Smith 1988b, 443). Not surprisingly, Norton suggests a
similar strategy; see Norton 1993,417.
116 NEWTON DA COSTA AND STEVEN FRENCH
guides to the completed successor. Our intention is not to dispute the rich historical
analysis which Smith presents in support of these claims. Certainly this process of
'projection' helps us to understand the way in which consistent theories come to be
constructed from their inconsistent predecessors. 21
However, we balk at the implication that such incomplete 'fragments' effectively
stood in for the inconsistent proposals and that scientists like Einstein and Bohr ' ...
made deductions not from an inconsistent but from an incomplete theory, only parts
of which were interpreted' (Smith 1988b, 443; his emphasis). It is true that they
speculated as to which well-confirmed fragments of the old quantum theory would
be retained by any future theory but what they reasoned from and drew inferences
from were the inconsistent models or theories themselves. Structurally these were
both inconsistent and incomplete in that with the inconsistency we do not know
whether the relevant properties and relations hold in the domain or not;22 as Post
says, the points where a theory is anomalous are also those where it is open. And of
course this openness or lack of closure (this time of the theory itself as viewed from
our extrinsic perspective) is representable model-theoretically by a partial structure.
Returning to the epistemic stance, at the heart of Smith's account lies a view of
acceptance that we find simply unacceptable (!):
Acceptance of a set of statements warrants their unrestricted use in all reasoning
processes. At least one member of an inconsistent set of statements must be false. So,
without additional information, we cannot rationally assent to the unconditional use of
anyone of the statements from a set known to be inconsistent. [Smith 1988b, 432)
Leaving aside the observation that this characterisation of inconsistency begs the
question against paraconsistent logics/ 3 no scientific theory, past or present, receives
such warrant; there are always hedging conditions, acknowledgments of limitations,
approximations etc., which are manifestations of the scientists doxastic attitudes
towards the theory. And these attitudes are inadequately and incorrectly
characterised in terms of belief in the (correspondence) truth of the theory. Our
account is an attempt at the difficult trick of accommodating these weaker attitudes
while retaining the connection between acceptance and belief of some form. 24 We
prefer to stick to 'acceptance' rather than 'entertainment', since even in the case of
inconsistent theories there is still the element of commitment which we (and Brown)
take to be the hallmark of the former (witness Bohr's elaboration and defence of his
theory).
REFERENCES
Arruda, A. l. (1980), A Survey of Paraconsistent Logic. In Mathematical Logic in Latin America, A. l.
Arruda, R. Chuaqui and N. C. A. da Costa (eds.), North-Holland, pp. 1-41.
Arruda, A. l. (1989), Aspects of the Historical Development of Paraconsistent Logic. In Paraconsistent
Logic: Essays on the Inconsistent, G. Priest, R. Routley, and J. Norman (eds.), Munich: Philosophia,
1989, pp. 99-130.
Bartelborth, T. (1989), Is Bohr's Model of the Atom Inconsistent? In Philosophy oJthe Natural Sciences.
Proceedings oJthe 13th International Wittgenstein Symposium, P. Weingartner and G. Schurz (eds.),
HPT, pp. 220-223.
Bohr, N. (1972), Collected Works, Elsevier.
Brown, B. (1990), How to be Realistic about Inconsistency in Science. Studies in History and Philosophy
oJScience 21, 281-294.
Brown, B. (1992), Old Quantum Theory: A Paraconsistent Approach. PSA 1992, Vol. 2, Philosophy of
Science Association, pp. 397-411.
Bueno, O. (1997), Empirical Adequacy: A Partial Structures Approach. Studies in History and
Philosophy oj Science 28, 585-610.
da Costa, N. C. A.(1989), Logic and Pragmatic Truth. In Logic, Methodology and Philosophy oJScience
VIII, J. E. Fenstad et al. (eds.), Elsevier, pp. 247-261.
da Costa, N. C. A., O. Bueno, and S. French (1998), The Logic of Pragmatic Truth. The Journal oj
Philosophical Logic 27, 603-620.
da Costa, N. C. A. and S. French (1989a), Pragmatic Truth and the Logic of Induction. The British
JournalJor the Philosophy oJScience 40,333-356.
da Costa, N. C. A. and S. French (1989b), Critical Study of 'In Contradiction'. Philosophical Quarterly
39,498-501.
da Costa, N. C. A. and S. French (1990), The Model-Theoretic Approach in the Philosophy of Science.
Philosophy oj Science 57, 248-265.
da Costa, N. C. A. and S. French (1993a), Towards an Acceptable Theory of Acceptance. In
Correspondence, In variance and Heuristics, S. French and H. Kamminga (eds.), Dordrecht: Reidel,
pp.137-158.
da Costa, N. C. A. and S. French (1993b), A Model Theoretic Approach to 'Natural Reasoning'.
International Studies in Philosophy oJScience 7,177-190.
da Costa, N. C. A. and D. Marconi (1989), An Overview of Paracon is tent Logic in the '80's. The Journal
oj Non-Classical Logic 6, 5-31.
French, S. (1991), Rationality, Consistency and Truth. The Journal oj Non-Classical Logic 7, 51-71.
118 NEWTON DA COSTA AND STEVEN FRENCH
French, S. and J. Ladyman (1997), Superconductivity and Structures: Revisiting The London Approach.
Studies in History and Philosophy of Modern Physics 28, 363-393.
French, S. and J. Ladyman (1999), Reinflating the Semantic Approach. International Studies in the
Philosophy ofScienc 13,103-121.
Gotesky, R. (1968), The Uses ofInconsistency. Philosophy and Phenomenological Research 28,
471-500.
Hettema, H. (1995), Bohr's Theory of the Atom 1913-1923: A Case Study in the Progress of Scientific
Research Programmes. Studies in History and Philosophy ofModern Physics 26, 307-323.
Jammer, M. (1966), The Conceptual Development of Quantum Mechanics, McGraw-Hill.
Klein, M. (1970), Paul Ehrenfest, Vol. I, North-Holland.
Lakatos, I. (1970), Falsification and the Methodology of Scientific Research Programs. In Criticism and
the Growth of Knowledge, I. Lakatos and A. Musgrave (eds.), Cambridge: Cambridge University
Press, pp. 91-195.
Laymon, R. (1985), Idealizations and the Testing of Theories by Experimentation. In Experiment and
Observation in Modern Science, P. Achinstein and O. Hannaway (eds.), Boston: MIT Press,
pp.147-173.
Laymon, R. (1988), The Michelson-Morley Experiment and the Appraisal of Theories. In Scrutinizing
Science, A. Donovan et al. (eds.), Dordrecht: Kluwer, 1988, pp. 245-266.
Madan, D. B. (1983), Inconsistent Theories as Scientific Objectives. Philosophy ofScience 50, 453-470.
Mikenberg, I., N. C. A. da Costa, and R. Chuaqui (1986), Pragmatic Truth and Approximation to Truth.
Journal of Symbolic Logic 51,201-221.
Norton, J. (1987), The Logical Inconsistency of the Old Quantum Theory of Black Body Radiation.
Philosophy of Science 54, 327-350.
Norton, J. (1993), A Paradox in Newtonian Gravitation Theory. PSA 1992 Vol. 2, East Lansing:
Philosophy of Science Association, pp. 412-420.
Pais, A. (1991), Niels Bohr's Times, Oxford: Oxford University Press.
Popper, K, (1940), What is Dialectic? Mind 49,403-426.
Popper, K. (1972), The Logic of Scientific Discovery, Hutchinson and Co. (1959).
Post, H. R. (1971), Correspondence, Invariance and Heuristics. Studies in History and Philosophy of
Science 2, 213-255.
Priest, G. (1987), In Contradiction. Martinus Nijhoff.
Priest, G. and R. Routley (1984), Introduction: Paraconsistent Logics. Studia Logica 43, 3-16.
Reichenbach, H. (l949), The Theory of Probability, University of California Press.
Smith, J. (l988a), Scientific reasoning or Damage Control: Alternative Proposals for Reasoning with
Inconsistent Representations of the World. PSA 1988 Vol. 1, East Lansing: Philosophy of Science
Association, pp. 241-248.
Smith, J. (I 988b), Inconsistency and Scientific Reasoning. Studies in History and Philosophy of Science
19,429-445.
Thompson, P. (1989), Explanation in the Semantic Conception of Theory Structure. PSA 1988 Vol. 2,
East Lansing: Philosophy of Science Association, pp. 286-296.
GRAHAM PRIEST
1. INTRODUCTION
What role does, or should, inconsistency play in the empirical sciences? This is the
question that I will address in this essay. The question is hardly a new one, but the
development of modem formal paraconsistent logics has a profound impact on the
subject. Paraconsistent logicians have realised that their subject has important
implications for the empirical sciences and the philosophy thereof, I but discussions
of the applications of paraconsistent logic have focused largely on non-empirical
areas, such as semantics and metaphysics. It therefore seems appropriate to address
the question directly.2
I will first address the issue of the specificity of the empirical sciences:
observation. Next, we will look at the role that inconsistency has played in science
(henceforth in this essay, I will often take the qualifier "empirical" for granted), and
the relation of this to paraconsistent logic. This will raise the question of how
inconsistent information ought to be treated in science: what criteria, for example,
may lead us to accepting an inconsistent theory? And should such an acceptance
ever be more than provisional? These topics will be addressed in the next sections.
An outcome of this discussion will be that, in the light of developments in
paraconsistent logic, we may well have to change our attitude to inconsistencies of
certain kinds; such a change would open whole new possibilities in science itself.
Before we can address the issue of the role of inconsistency in empirical sciences, it
will be important to discuss how such sciences differ from similar inquiries (which
are often intricately connected with them), such as mathematics and metaphysics.
The standard answer to this question is that in the empirical sciences, but not the
others, observation plays a role. In the more developed sciences such observations
I See, e.g., Priest and Routley 1989a, 367-79 and Priest and Routley 1989b, 494ff.
2 I am grateful to Joke Meheus for the promptings to do so, and for her insightful comments on an earlier
draft of the paper. I am also grateful to Diderik Batens for thought-provoking discussions on the topic
over the years, and for his comments on a draft of the paper. A version of the paper was given at a
seminar in the Philosophy Department at the University of Melbourne. I am grateful to many of the
participants there for their comments and questions.
119
J. Meheus (ed.), Inconsistency in Science, 119-128.
© 2002 Kluwer Academic Publishers.
120 GRAHAM PRIEST
are obtained through active experimentation. The observations serve to provide the
ultimate explananda for science, as well as providing important inputs into the
evaluation of scientific theories. This standard answer is not entirely unproblematic:
the role of empirical data in science has often been overrated, and its role in areas
such as philosophy under-rated-for (:xample, by positivists and empiricists.
However, that it plays a much more central role in science than in other inquiries can
hardly be denied.
The distinction between what is observable and what is not, is not, itself, entirely
unproblematic. Some states of affairs are certainly observable (such as the colour of
an insect) and some are certainly not (such as the colour of a quark). But it is
impossible to draw a sharp boundary between what is, and what is not. What is
observable may depend on what aids to perception (such as a microscope) are used;
or on how we interpret our sensory input in the light of accepted theories (such as
theories that tell us that when we witness the tracks in a bubble-chamber, we are
seeing the tracks of charged particles). However, as with all vague or context-
dependant distinctions, the fact that it is impossible to draw a neat line between its
two sides does not mean that there is not an important distinction to be drawn.
(Compare: being a child and being an adult.)
We can now proceed to an issue of importance in what follows. What can one
see?3 For a start, seeing, in the sense relevant to explanation and confirmation, is
propositional. We see that something is the case. We see that the stars appear to be
in a certain position and ask for an explanation, or use this against a theory
according to which they should appear somewhere else.
Can we also see that something is not the case? Some have thought not. 4 We
always see that something is the case, and then infer that something else is not the
case. For example, we see that something is red and infer that it is not black. Whilst
one may certainly do this, it seems to me that one need not: one can see directly that
something is not the case. Try a thought experiment. I show you an ordinary apple
and ask: is this black? You compare its colour with a mental paradigm of blackness,
and it does not match; you say no.5 Or again, you enter a room; the whole room is
visible from where you stand; there is no one there. You can see that Pierre is not in
the room. You do not have to say: the things in the room are a chair, a table ... ;
Pierre is not a chair, Pierre is not a table ... ; therefore, etc. 6 Even the very distinction
between seeing what is the case and what is not the case is a false one. Some seeings
are both. When talking of physical objects, to be transparent is not to be opaque and
vice versa. But you can see that something is transparent and you can see that
something is opaque.
It should be remembered that seeing is not simply a matter of light rays hitting
the retina. Certainly, the eyes are involved in seeing; but one needs more than eyes
to see. If what one sees is to playa role in one's cognitive functioning, one must also
3 In what follows, I will restrict myself to discussing vision, since this is by far the most important
sensory modality in science; but similar comments apply to the other senses.
4 E.g., Vasil'ev (1913).
5 I am not suggesting that all vision is paradigm-based in this way, as should become clear in a moment.
6 The example comes from Sartre (1943, ch.l, sect.2).
INCONSISTENCY AND THE EMPIRICAL SCIENCES 121
understand one's visual input. Hence, the categories of the understanding playa
role. To see is to interpret ones visual stimuli, by applying these categories, either
consciously or preconsciously; and there is no reason why truth functions such as
negation or disjunction should not enter into the process of interpretation directly.
For example, I can see that this is a photograph of either Ned or Ted, identical twins
such that I cannot tell the difference, without seeing that it is a photograph of Ned,
and inferring the disjunction.
Next step: if u and ~ are states of affairs observable at the same time and place,7
then so is their conjunction. We can see that something is a unicorn; we can see that
it is green. Hence, we can see that it is a green unicorn. Applying this: if u and 'u
are observable states of affairs, so is u 1\ 'U. Of course, this is not to say that the
conjunction is observed; merely that it is observable-that is, it is of a kind such that
if it were to be the case, it could be seen.
One might doubt this. Might it not be the case that our cognitive functioning
makes it impossible for us to see an inconsistent state of affairs? Might not the way
our perception works impose a 'consistency filter' on what we see? No. Seeing
impossible situations is quite possible. This is what we perceive in various visual
illusions. Thus, for example, there are many well known impossible figures (of the
kind, for example, employed by Escher in his drawings); there are perceptual sets
where people report seeing things as simultaneously red and green; there are
situations where things appear to be moving and not moving. Let me just describe
one of these in more detail, the last. 8 This is commonly known as the waterfall
illusion. After conditioning the visual field by showing it constant motion of a
certain kind, say a spinning spiral, one then looks at a stationary scene. The scene
appears to move in the opposite direction, but nothing in the scene changes its
position; for example, an object at the top of the visual field does not move round to
the bottom. What we see appears to be both moving and stationary; this is the
natural way of describing one's visual sensations. Of course, perception in the cases
I have described is not veridical;9 these are illusions, and things are not really thus;
but that is how they would appear if things were thus. There is therefore nothing
about our visual system that requires perception to be consistent.
3. TYPES OF INCONSISTENCY
7 From the same perspective-and whatever other qualifications one needs to make to rule out irrelevant
counter-examples.
8 Details of examples such as this can be found in most books on vision and visual illusions, e.g.,
Robinson 1972. Examples of this kind are further discussed in Feyerabend 1975, 258ff., and Priest
1999a.
9 Which is not to say that a perception of a contradictory state of affairs cannot be veridical. But then,
how do we know that the cases in point are illusions? Many things tell us this. For example, in the
case of the waterfall illusion, the apparent motion of, say, the room, is not confirmed by our other
senses, or by other people. And if the room were really moving, things would have a tendency fall
over, which they do not, etc.
122 GRAHAM PRIEST
historians and philosophers of science have observed that there are such
inconsistencies-indeed that they are common-even in contemporary science. 10 If
we distinguish between observation and theory (what cannot be observed), then
three different types of contradiction are particularly noteworthy for our purposes:
between theory and observation, between theory and theory, and internal to a theory
itself. 11 Let us look at these three in more detail.
Inconsistency between theory and observation is the most obvious example. We
have a well-received theory, T, with a certain observable consequence, u. We then
run an experiment and observe that 'u. Simple-minded falsificationists would
suggest that this shows that T is wrong and is to be ditched. But many philosophers
of science have pointed out that this does not necessarily happen. 12 The contradiction
may be treated as the site of an anomaly: both T and 'u may be accepted pro tern. T
will not be jettisoned until we have a better alternative; 'u will not be jettisoned
until we have an explanation of why our observation was incorrect. Examples of this
kind in the history of science are legion. The precession of the perihelion of
Mercury, at odds with Newtonian celestial dynamics, was known for a long time
before the rejection of Newtonian dynamics in favour of special relativity. Prout's
hypothesis was widely accepted by many chemists even though it was known to be
at odds with empirical data (much of which was subsequently rejected). And so on.
The second kind of inconsistency, between theory and theory, is less frequently
noted, but certainly occurs. This is when we have two well-accepted theories, TJ and
T2 , which have inconsistent consequences. Again, though this may be noted as the
site of a problem, both TJ and T2 are retained until a suitable replacement theory for
one or both is found. An example of this concerns the age of the earth in late 19 th
century science. According to evolutionary theory (which was, by that time, well
accepted), the age of the earth had to be hundreds of millions of years; but according
to received thermodynamics, the sun-and so the earth-could not possibly be that
old. (The issue was resolved only in the 20 th century with the discovery of
radioactivity, a hitherto unsuspected form of heat-generation.) Another,
contemporary, example: the theories of relativity and quantum theory are known to
be mutually inconsistent.
The third example of an inconsistency is when a theory is self-inconsistent. This
could arise because a theory has inconsistent observational consequences, though I
know of no interesting cases of this in the history of science. What there certainly
have been are inconsistent theories where the inconsistencies are located internally,
away from observational consequences. For example, for over a hundred years
Newtonian dynamics was based on the old calculus, in which infinitesimals had
inconsistent properties (being both non-zero, at one point in computations, and zero
at others). Another, particularly striking, example is Bohr's theory of the atom,
which included both classical electrodynamic principles and quantum principles that
10 For example, Lakatos 1970, Feyerabend 1975, ch.5, which can be consulted for details of many of the
following examples.
11 A good general discussion of kinds of problem-situations in science, including those involving various
kinds of contradictions, can be found in Laudan 1977, chs.l, 2.
12 E.g., Kuhn (1970), Lakatos (1970), and Feyerabend (1975).
INCONSISTENCY AND THE EMPIRICAL SCIENCES 123
were quite inconsistent with them. Though the theory was certainly considered as
problematic, its empirical predications were so much better than those of any other
theory at the time that it had no real competitor. 13
4. HANDLING INCONSISTENCY
As we have seen, the corpus of scientific beliefs may, or is even likely to be,
inconsistent at any time. But from things accepted, scientists infer other things that
they accept, and they do not infer arbitrary conclusions. It follows that the inference
procedure employed here must be a paraconsistent one (where an arbitrary a and °a
do not entail an arbitrary ~). What paraconsistent inference procedure is employed in
inconsistent cases is another question; there is no a priori reason to suppose that it
must be one of the standard monotonic paraconsistent logics, or even that it must be
monotonic at all. Nor should one suppose that it must be the same in every case.
What is guaranteed is that there must be some systematic procedure for drawing
conclusions, and which does not permit drawing an arbitrary conclusion from a
contradiction. What procedure is or was employed in any given case is a matter for
detailed investigation.
I do not intend to discuss detailed examples here; but for what follows, it will be
important for us to distinguish between two different kinds of paraconsistent logic:
adjunctive and non-adjunctive. In adjunctive paraconsistent logics, such as standard
relevant logics and da Costa's C systems, the rule of adjunction, a, ~ 1= a 1\ ~, is
valid. In non-adjunctive systems, such as discussive logics, it is not. Non-adjunctive
paraconsistent logics often employ some chunking procedure. Because of the failure
of adjunction, one cannot simply put arbitrary premises together and draw
conclusions. If the inference procedure is not to be impoverished, it is usually
necessary to be able to put some premises together (into a chunk). A simple
procedure is to put together any bunch of premises that are mutually consistent; but
there are more sophisticated ones. 14
One thing that is clear from a fairly cursory consideration of the scientific
handling of inconsistent information is that the inference procedure employed is
often of a non-adjunctive, chunking, variety. For example, given the dispute about
the age of the earth at the end of the 19th century, no one conjoined the views that
the earth was hundreds of millions of years old, and that it was not, to infer that the
13 It might be suggested that in the situations mentioned, the scientific community did not really accept
inconsistencies. Rather, they had degrees of belief in various propositions, and their degrees of belief
in the contradictory a and -'a were less than I. Now, it may well be the case that there are degrees of
belief, and that acceptance is to be understood as having a sufficiently high degree; but this suggestion
will not really help. For example, given the inconsistency between the theory of evolution and
thermodynamics, it would follow that one or both of these theories was believed to degree ~ 0.5,
which is to say that one or both of these was not accepted at all, which is untrue. Or if both were
accepted to degree 0.5, and one sets the level acceptance at this figure, then inconsistencies were
accepted, as claimed.
14 For a general account of paraconsistent logic, see Priest 2002. On non-adjunctive strategies, see,
especially, section 4.2.
124 GRAHAM PRIEST
earth really had a contradictory age. IS Similarly, in the Bohr theory of the atom, the
drawing of conclusions was restricted to well-defined consistent chunks, possibly in
accordance with some pragmatically determined-but still determinate-
considerations. 16 A conclusion drawn may then have been fed into another consistent
chunk which contained information inconsistent with that in the first chunk. 17
Whether adjunctive paraconsistent logics have, historically, ever been used in
handling inconsistencies is a different matter, and is rather doubtful. At any rate, I
know of no examples where this is clearly the case.
We have seen that inconsistent information has sometimes been accepted in the
history of science; we have also seen, at least in outline, why this does not lead to
disaster. None of this shows that the inconsistent information ought to have been
accepted (even provisionally). But the situation seems to have arisen so frequently
that it is implausible to level at the scientific community charges of blatant and
frequent irrationality. This, therefore, raises the question of the conditions under
which it is reasonable to accept an inconsistent theory or other body of information.
The question of what makes it reasonable to accept any theory is a familiar one
in contemporary philosophy of science, and a very difficult one. Neither do I intend
to advance a detailed answer to that question here. One thing that is, I think, fairly
universally agreed is that the doxastic goodness of a theory may be evaluated under
a number of orthogonal criteria. Let met explain.
A major criterion is empirical adequacy. For any observational consequence, (1,
that a theory entails, if (1 is observed, this is a good mark. If it is not, this is a black
mark. Empirical adequacy is, perhaps, the most important criterion in science. It is,
at least, the one that mainly distinguishes the empirical sciences from similar
investigations. But it is certainly not the only one; nor can it be. It is at least
theoretically possible to have different theories that are both empirically adequate;
more commonly, it happens that no theory in the field is entirely empirically
adequate. Hence other criteria have also to be employed. There is a multitude of
these;18 philosophers may disagree both about what they are, and about how, exactly,
to understand them, but the following have certainly been suggested, and are very
plausible. Good-making features include: simplicity, ontological leanness
(Ockham's razor), explanatory power, a low degree of ad hocness, unity,
fruitfulness. The converse features are bad: complexity, ontological extravagance,
explanatory poverty, much ad hocness, fragmentation, barrenness. I have not
15 Which is not to say that people did not notice that the theories together entail something of the form
U /\ -'U, and so conclude that there was a problem here; merely that the corpus of accepted beliefs was
not closed under an adjunctive logic.
16 Many non-monotonic (paraconsistent) logics incorporate pragmatic features of this kind; for example,
in the ordering relation with respect to which minimisation is defined. See Priest 1999b.
17 For a more detailed discussion, see Smith 1988, Brown 1990 and, especially, Brown 1993. See also the
discussion of the inconsistent early quantum theory of black body radiation in Norton 1989 (esp.
pp.330ff.).
18 See, e.g., Quine and Ullian 1970, ch.5, Kuhn 1977, and Lycan 1988, ch.7.
INCONSISTENCY AND THE EMPIRICAL SCIENCES 125
7. INCONSISTENT MATHEMATICS
20 More generally, if the inconsistency is handled with an inference mechanism that does not respect
truth-preservation, the same conclusion follows.
21 The Law of Non-Contradiction is taken on in Priest 1987. A number of other arguments for the Law
are discussed and rejected in Priest 1998.
22 This raises the following question. Suppose that we have a theory based on a non-adjunctive logic.
This, as I have argued, is not an ultimately acceptable theory. Why can we not tum it into one, simply
by changing the underlying logic to an adjunctive (truth-tracking) one? This may be a possibility,
though not if changing the logic results in the theory being empirically inadequate-which is
normally why a non-adjunctive procedure is used in the first place.
INCONSISTENCY AND THE EMPIRICAL SCIENCES 127
law has rarely been challenged. 23 Hence, scientists and philosophers have not been
prepared to brook the thought that an inconsistent theory of any kind might be true.
But subscribing to the Law is not rationally mandatory, as the development of
paraconsistent logics has played a large role in showing. Once this fact is digested,
scientists may well-justifiably-take a different attitude to inconsistent theories of
the appropriate kind. Indeed, they may even develop inconsistent theories, if these
have the right empirical consequences, just as paraconsistent logicians have
articulated inconsistent theories of semantics to handle the paradoxes of self-
reference.
In modern science, the inferentially sophisticated part is nearly always
mathematical. An appropriate mathematical theory is found, and its theorems are
applied. Hence, a likely way for an inconsistent theory to arise now in science is via
the application of an inconsistent mathematical theory. Though the construction of
inconsistent mathematical theories (based on adjunctive paraconsistent logics) is
relatively new, there are already a number: inconsistent number theories, linear
algebras, category theories; and it is clear that there is much more scope in this
area. 24 These theories have not been developed with an eye to their applicability in
science-just as classical group-theory was not. But once the paraconsistent
revolution has been digested, it is by no means implausible to suppose that these
theories, or ones like them, may find physical application-just as group-theory did.
For example, we might determine that certain physical magnitudes appear to be
governed by the laws of some inconsistent arithmetic, where, for example, if nand
m are magnitudes no smaller than some constant k, n + m = k (as well as its being
the case that n + m =1= k).25 There are, after all, plenty of episodes in the history of
science in which we came to accept that certain physical magnitudes have somewhat
surprising mathematical properties (being imaginary, non-commuting, etc.). Why
not inconsistency? Which is not to say that an inconsistent mathematical theory must
be interpreted realistically. Such theories may have instrumental uses, just as much
as consistent theories.
8. CONCLUSION
I believe that the development of modern formal paraconsistent logics is one of the
most significant intellectual developments of the 20th century. In challenging
entrenched Western attitudes to inconsistency that are over 2,000 years old, it has
the potential to ricochet through all our intellectual life-and empirical science
wears no bullet-proof vest. As we have seen, empirical scientists have always
23 An analysis of Aristotle's arguments for the Law of Non-Contradiction can be found in Priest 1997.
24 For inconsistent arithmetic, see Priest 2002, section 9. On inconsistent mathematics in general, see
Mortensen 1995.
25 For a thought experiment illustrating how this might come about, see Priest 200b. There is even one
place where an inconsistent mathematics might possibly find an application already. In the two-slit
experiment in quantum mechanics, the causal anomaly can be resolved by supposing that the photon
does the impossible, going through the two slits simultaneously, and handling this with an adjunctive
paraconsistent probability theory. For details, see Priest and Routley 1989a, 377ff.
128 GRAHAM PRIEST
tolerated, and operated within, inconsistency in certain ways. One of the liberating
effects of paraconsistency should be to allow us to understand better exactly how
this proceeded. Such an understanding is bound to reflect into our understanding of
the rationality of theory-choice, in the ways that I have indicated. Perhaps most
importantly of all, paraconsistency may open the gate to important new kinds of
theory within science itself. Where this will all lead, one cannot even begin to
speculate.
REFERENCES
Brown, B. (1990), How to be a Realist about Inconsistency in Science. Studies in History and Philosophy
of Science 21,281-94.
Brown, B. (1993), Old Quantum Theory: a Paraconsistent Approach. Proceedings of the Philosophy of
Science Association, Vol. 2, 397-441.
Feyerabend, P. (1975), Against Method. London: New Left Books.
Kuhn, T. S. (1962), The Structure of Scientific Revolutions. Chicago: University of Chicago Press.
Kuhn, T. S. (1977), Objectivity, Value Judgment and Theory Choice. ch.13 of The Essential Tension,
Chicago: University of Chicago Press.
Lakatos, I. (1970), Falsification and the Methodology of Scientific Research Programmes. In Criticism
and the Growth of Scientific Knowledge, I. Lakatos and A. Musgrave (eds.), Cambridge: Cambridge
University Press, pp.91-196.
Laudan, L. (1977), Progress and its Problems, London: Routledge and Kegan Paul.
Lycan, W. (1988), Judgment and Justification. Cambridge: Cambridge University Press.
Mortensen, C. (1995), Inconsistent Mathematics. Dordrecht: Kluwer Academic Publishers.
Norton, J. (1987), The Logical Inconsistency of the Old Quantum Theory of Black Body Radiation.
Philosophy of Science 54,327-50.
Priest, G. (1987), In Contradiction. Dordrecht: Martinus Nijhoff.
Priest, G. (1997), To Be and not to Be-That is the Answer. Philosophiegeschichte und Logische Analyse
1,91-130.
Priest, G. (1998), What's So Bad About Contradictions? Journal of Philosophy 95, 410-26.
Priest, G. (1999a), Perceiving Contradictions. Australasian Journal of Philosophy 77, 439-46.
Priest, G. (1999b), Validity. What is Logic?, European Review of Philosophy 4, 183-206.
Priest, G. (2002), Paraconsistent Logic. In D. Gabbay and F. Guenthner (eds.), Handbook of
Philosophical Logic, second edition, vol. 7, Dordrecht: Kluwer Academic Publishers.
Priest, G. (200a), Paraconsistent Belief Revision. Theoria, forthcoming.
Priest, G. (200b), On Alternative Geometries, Arithmetics and Logics, a Tribute to Lukasiewicz. In
Proceedings of the Conference Lukasiewicz in Dublin, P. Simons (ed.), to appear.
Priest, G. and R. Routley (I 989a), Applications of Paraconsistent Logic. ch. 13 ofG. Priest et al. (1989).
Priest, G. and R. Routley (I 989b), The Philosophical Significance and Inevitability of Paraconsistency.
ch. 18 ofG. Priest et al. (1989).
Priest, G., R. Routley and J. Norman (1989), Paraconsistent Logic. Essays on the Inconsistent. Miinchen:
Philosophia Verlag.
Quine, W. V. and J. Ullian (1970), The Web of Belief Random House.
Robinson, J. O. (1972), The Psychology of Visual Il/usions. Hutchinson and Co.
Sartre, J.-P. (1943), L 'Etre et Neant. Gallimard.
Smith, J. M. (1988), Inconsistency and Scientific Reasoning. Studies in History and Philosophy of
Science 19,429-45
Vasil'ev, N. (1913), Logica i Metalogica. Logos 1-2, 53-81. Translated into English by V. Vasukov as
Logic and Metalogic. Axiomathes 4 (\993), 329-5\.
DID ERIK BATENS
Abstract: This paper states and defends the philosophical programme underlying the Ghent approach to
adaptive logics. Two central arguments are epistemic in nature, one logical. The underlying claim is that
even people with rather classical views should see adaptive logics as the only sensible way to handle the
inconsistencies that regularly arise in human knowledge, including scientific theories.
The main application context that led to the first adaptive logics, was a specific kind
of creative l problem solving processes in the presence of inconsistencies. Consider a
theory T = (f, L) and let OtLf = {A If I-L A}, the deductive closure, defined by
the logic L, of the set of (non-logical) axioms f. If T was intended as consistent, L
will presumably be Classical Logic 2-henceforth CL. Sometimes L turns out to be
inadequate with respect to T in that f violates some of the presuppositions of L. If L
is CL, the inadequacy always surfaces in the form of an inconsistency, and hence of
the triviality of Ot df).3 Usually, there are good reasons to search for a consistent
replacement of f (rather than for a replacement of L). However, especially if T is a
valuable theory, one will look for a way to provisionally live with T. Moreover, in
search for a replacement, one will not start from scratch, but rather reason from T.
How T should be interpreted in such cases, was the problem that led to
inconsistency-adaptive logics.
Meanwhile the adaptive logics enterprise led to a host of technical results and the
logics proved to have many application contexts. In the present paper, I defend the
programme that is behind my approach to these logics. First, however, three
warnings that should be issued.
* Research for this paper was supported by Ghent University, by the Fund for Scientific Research -
Flanders, and indirectly by the Flemish Minister responsible for Science and Technology (contract
BIL98173). 1 am indebted to Graham Priest for some useful comments on a previous version of this
paper.
1 See Meheus 1997 for a sensible approach to the creativity of a problem solving process.
2 This claim is argued for in Section 5.
3 Examples from the history of mathematics are well known: Cantor's set theory, Frege's set theory,
Newton's infinitesimal calculus, etc. For case studies of inconsistencies in empirical theories, see
Norton 1987, 1993, Smith 1988, Nersessian 2002, and Meheus 1993,200+.
129
J. Meheus (ed.), Inconsistency in Science, 129-150.
© 2002 Kluwer Academic Publishers.
130 DID ERIK BA TENS
The first concerns other uses of adaptive logics. Some such logics have
interesting applications to inconsistent arithmetics; some allow one to reconstruct
non-monotonic logics popular in artificial intelligence; some offer sensible formal
approaches to ambiguity, to incompleteness, and to a host of other logical
abnormalities (the falsehood of A in the presence of A&B, etc.). In general, adaptive
logics present an outlook at a formal approach to argumentation contexts, and to
reasoning in natural languages. All this will not be discussed here. It is not, however,
irrelevant to the problems I shall be dealing with. When scientists have to leave the
narrow domain of algorithmic safety-the presence of an inconsistency is felt by
most as taking them outside that domain-their reasoning is typically of the same
nature as argumentation and, in general, as reasoning in natural language (even if the
scientists' reasoning continues to involve complex mathematical technicalities).
As other theories, formal systems may be applied in contexts for which they
were not originally intended. Moreover, they may be incorporated into a research
tradition that is radically different from the one in which they originated. Adaptive
logics are not an exception. In his 1991, Graham Priest has adapted the ideas
underlying inconsistency-adaptive logics to his dialetheic viewpoint. My approach
derives from specific views on knowledge, natural languages, conceptual systems,
and conceptual dynamics. Priest disagrees with most of these views. A reader that is
not convinced by the arguments I shall provide in favour of my underlying views,
should not conclude from this that adaptive logics are useless or uninteresting. This
was the second warning.
The third warning concerns the force of inconsistency-adaptive logics. Unlike
classical logic, they prevent inconsistencies from leading to triviality. Unlike
monotonic paraconsistent logics, 4 they interpret an inconsistent set of premises (and
hence also a theory) as consistently as possible (see Section 4). So, they enable us to
live with inconsistencies, while (on my view) still classifying them as problematic-
see again Section 4. However, inconsistency-adaptive logics do not resolve any
inconsistencies, and were never intended to do so. It is not difficult to find some
inconsistency-reducing mechanism 5 and to incorporate it into an inconsistency-
adaptive logic. The resulting logic, however, would be of limited use. The way in
which inconsistencies are removed should rely on non-logical (empirical,
methodological, or conceptual) considerations. Of course, some logical
inconsistency-elimination strategy might pick the correct alternative in a specific
case, but even if it does, the justification for the choice is non-logical.
To defend my programme I first present two central epistemological views
(Sections 2 and 3). Then I briefly show the way in which adaptive logics agree with
them (Section 4). The following two Sections (5 and 6) concern more specific
4 A logic is monotonic if and only if everything derivable from r is derivable from rut!.. A logic is
paraconsistent if and only if it does not validate A, -A r- B, the so-called Ex Fa/so Quodlibet
(sometimes called Ex Contradictione Quodlibet or Explosion). In other words, a paraconsistent logic
does not turn all inconsistent theories into trivial theories.
5 Here is a simple one: if A is true in all minimally inconsistent models of the premises-see below-and
-A is not (or vice versa), eliminate -A (or vice versa); otherwise eliminate both A and -A. An
interesting list of (logical and other) inconsistency-reducing mechanisms can be found in Darden
1991.
IN DEFENCE OF A PROGRAMME FOR HANDLING INCONSISTENCIES 131
choices among adaptive logics. The first of them contains a rather lengthy defence
of CL against attacks by relevantists and dialetheists. In Section 7, I list some
perspectives and open problems.
2. WHY CONSISTENCY?
6 Zeno's paradoxes nicely illustrate the point: in some conceptual systems, movement necessarily
involves inconsistency.
7 To give just one example of a typical problem: any language L that, in combination with some R, allows
for a consistent description of even some subsystem of the world, might require more than a
denumerable number of individual constants, or might require predicates of a non-enumerable rank.
Handling such theories for the prediction or explanation of facts is beyond human capacities.
132 DIDERIK BATENS
superior to all its alternatives in other respects. Moreover, other problems may be by
far more urgent than the lack of determinism. Finally, the fact that determinism is a
justified methodological requirement does not entail that a fully deterministic true
theory about the domain is possible, and even less that such a theory is within the
reach of humans. All this holds just as well for consistency.
Some people seem to think that there are standard means to restore consistency.
Suppose that, according to some theory, some objects have property P, others have
property ~ P, and still others have property P as well as ~ P. It is obviously possible
to modify the language L to a language L' in such a way that objects of the three
sorts are respectively characterized in L' by three predicates, PI. P 2, and P 3 that form
a 'family' (are mutually exclusive and collectively exhaustive). With this move, the
theory may become consistent. And one might argue that the replacement of Pa by
PI aV P 2 a does not necessarily decrease the precision of the theory. However, a
serious problem may (and usually will) arrive: that the theory loses in coherence and
conceptual clarity. So, the upshot may (and usually will) be problematic from a
conceptual point of view, and hence may be worse than the inconsistent theory.
Moreover, the theory obtained may very well be still inconsistent, as a simple
example illustrates. Suppose one tried to eliminate the Russell paradox by replacing
the membership relation (E) by a family of three relations, where x ElY corresponds
to the situation in which only x E Y is true, x E2Y corresponds to the situation in
which both x E Y and ~x E yare true, and x E3Y corresponds to the situation in
which only ~x E Y is true. But whether the abstraction axiom is formulated as
C3x)(Vy)(y Elx == A(y)) or as (3x)(Vy)((y Elx V Y E2X) == A(y)), contradictions are
still easily derived.
The situation is different for inconsistencies that derive from observational
criteria (including those that refer to instruments). As, for many properties, we rely
upon a multiplicity of such criteria-several (kinds of) thermometers, etc.-
inconsistencies may be easily generated. If an inconsistency is produced by the
application of two criteria for the same property (or fact, etc.), we may easily
disconnect one or more criteria-"this thermometer measures 20 whereas the other
measures 25" is obviously not a contradiction. We still have a problem, viz. which
(if any) thermometer is an adequate measuring device for temperature. (Or, why the
two thermometers give a different outcome.) Yet, it is quite obvious that this is the
real problem, whereas the inconsistency is not. Could the inconsistency derive from
a single measuring instrument (or other observational criterium)? I honestly cannot
imagine any example of this, except where the criterium refers to some vague
property (as in "it rains" or "he is bald"). But if a vague predicate occurs in the
language of some scientific theory, everyone will readily recognize that the real
problem lies there, and not with the possibly resulting inconsistency. If I am right,
inconsistencies that occur from observations are easily removed at the expense of
facing the problem that generated them. As I argued before, the situation is quite
different if an inconsistency derives from some theory. In that case, it may be
preferable to retain the inconsistency rather than to create a different and more
annoying problem.
Does it not follow that the world is necessarily consistent provided we interpret
'the world' in the way Mach (and other positivists) did, viz. as the set of the
IN DEFENCE OF A PROGRAMME FOR HANDLING INCONSISTENCIES 133
phenomena (the set of our observational data)? The answer is still no. What we
actually did experience cannot sensibly be identified with 'the world'. The very least
we want to include is what is observable. Apart from all kinds of problems
connected to this category-remember the discussion on van Fraassen's attempt at
defining it-we justify most of our beliefs about what is observable by relying on
theories. 8
Let me sum up and add a final warning. Consistency is a methodological
requirement. Our best knowledge may be inconsistent, and may remain so forever.
The need for paraconsistent logics follows immediately. But what about inconsistent
mathematical systems? In all domains of mathematics, we have theories that we
consider as consistent (even if an absolute warrant is not available). Some might
derive from my above arguments that it is perverse to study systems that we know to
be inconsistent. If consistency is a methodological requirement, they might reason, it
cannot be justified to study inconsistent theories for which (presumably) consistent
alternatives are available. This argument is mistaken. Inconsistent arithmetic
systems may be required by other inconsistent theories. Moreover, their study may
offer invaluable metatheoretic insights.
8 I do not believe in the possibility of a strict distinction between what is observable and what is not. This,
however, does not undermine the argument from the previous paragraph: if an inconsistency arises
from observation alone, we may easily, and justifiably, remove the inconsistency and face the (real)
problem that affects our observational criteria.
134 DIDERIK BA TENS
9 An explanation for the confusion may be found in the strong influence of positivist epistemology at the
time the combinatorial technique was transposed from formal to natural languages. Mach and others
required that all 'interpretations' were eliminated from scientific theories (although not from their
genesis). So, in those days, the basic meaning entities were supposed to be free of presuppositions.
IN DEFENCE OF A PROGRAMME FOR HANDLING INCONSISTENCIES 135
\0 Actually, the situation is worse. Many sentences uttered by scientists refer to non-existing entities (see
again Laudan 1981). Which proposition, if any, is expressed by such sentences on the fact-of-the-
matter view of propositions? In view of this, what is derivable, e.g., from Maxwell's theory?
\I Often, an addressee (or third party) understands something different from a speaker's intentions.
Sometimes we become aware that one of our utterances may be understood in a different (perhaps
more interesting) way. Sometimes we discover presuppositions of our utterances. All this is familiar
and straightforward, and does not require any meanings that transcend our or other people's
understanding.
136 DIDERIK BATENS
is directly relevant for the flexibility of our thinking. Precisely this flexibility enables
us to understand others, to vary conceptual frames, and thus to arrive at novel views
and theories. Sometimes we consciously decide not to take into account part of our
knowledge. That Clausius was able to reach his 1850 remarkable synthesis
(including the very first version of the entropy principle), was largely due to his
ability to think about heat as a whatever-it-is that is transferred in the Camot cycle
and is 'produced' when one turns the handle of Joule's electricity machines-see
Meheus (1993, §5.4).
Some approaches to decision making, creativity and discovery rely on such
mechanisms as expansion and contraction. This road leads nowhere. It still relies on
the standard and mistaken conception: that the totality of our knowledge system is
involved in such processes, and hence is expanded and contracted. 12 An easier and
much more convincing explanation provides from the flexibility of contexts.
What are the consequences of all this for logic? Simply that the correctness of
reasoning depends on the specific conceptual system in which it occurs. Phrased
differently, it depends on properties of the context (more precisely on the contextual
certainties). These properties do not depend on arbitrary decisions. 13 So, my
contextual approach does not involve any instrumentalism or relativism. I profess
objectivism in logical matters. Only, the objectivity is neither determined by the
world (the facts-of-the-matter) nor by anything like the common features of all
possible contexts (which are presumably nil anyway). Rather, the objectivity refers
to specific properties of the context in which our reasoning proceeds.
In some contexts the consistency of the domain functions as a certainty. In
others, the relevant statements are patently inconsistent. Some contexts require truth-
functional connectives only, while others involve relevant implications. As a result,
the logics employed in different contexts may be fully 'incompatible'. People with
some relevant training are able to reason in terms of classical logic with respect to,
say, GOdel's theorem, in terms of intuitionistic logic with respect to Brouwer's
choice sequences, and in terms of R with respect to Bob Meyer's relevant
arithmetic. They are able to so independently of their own view on mathematics. As
I said before, precisely this flexibility enables us to understand others, and to be
creative and inventive.
Adaptive logics are typically suited for contexts in which inconsistencies (or
other abnormalities) occur but are considered either as exceptional or as
problematic-different applications are discussed in Vermeir 200+, Batens 1994,
and elsewhere. Their status is not that of a pseudo-logic or quasi-logic. Just as any
other logic, they are suited for specific contexts (and not suited for others).
12 As I argued in Batens 1985b this mechanism fails to account for some obvious and simple
epistemological moves.
13 Those properties depend on our knowledge system and on rational decisions arrived at from and in
view of our knowledge system-see Batens 1985b, and 1992a.
IN DEFENCE OF A PROGRAMME FOR HANDLING INCONSISTENCIES 137
4. INCONSISTENCY-ADAPTIVE LOGICS
14 I put the expression between quotes because adaptive logics differ amongst each other by the way in
which they interpret the expression.
15 This is quite different from 'provided the involved part of the theory is normal'. I refer to the sentences
that behave normally according to the set of premises, not to the normal subsets of the premises-but
see Section 6.
16 The Ghent naming policy for adaptive logics: "A" + the name of the lower limit logic + a digit
referring to the strategy-"2" refers to minimal abnormality.
138 DIDERIK BA TENS
17 Priest here opts for his preferred paraconsistent logic LP, in which there is no detachable implication
(that is: no implication for which Modus Ponens is definable). In his 1987, a strict implication is
added to the system,
18 See Sections 1 and 7 on adaptive logics that are not inconsistency-adaptive.
19 The same holds for da Costa's C n systems (for n < (0).
20 For more than twenty-five years, I have been teaching my logic students (several thousands by now) a
simple and manageable relevant implication (and other non-standard materials), next to CL.
21 Or, obviously, in terms of an inconsistency-adaptive logic that has CL as its upper limit logic,
IN DEFENCE OF A PROGRAMME FOR HANDLING INCONSISTENCIES 139
they are not terribly interested. If one talks to them about relevance, let alone
inconsistency, they walk away.
Logicians did a bad job here. There are some results on inconsistent and relevant
mathematics; some more are forthcoming. But there are nearly no results in the other
sciences. Why would scientists tum to more complex logical systems if their need is
not spelled out to them? One might even question whether the available non-
classical results are important to scientists. Take Relevant Arithmetic as an example.
It has the impressive advantage over Classical Arithmetic, that (in several although
not all senses) its consistency can be proved by 'finitistic' means (one may prove in
terms of a finite model that 0 = 1 is not a theorem). However, the very fact that it
has finite models, shows that Relevant Arithmetic does not require the existence of
an infinite sequence of natural numbers. Few mathematicians will swallow this. 22
Let me return to the main point. Although it is easily seen that scientists frequently
use relevant implications (and other non-troth-functional logical constants) in
reasoning about their theories, the theories themselves are commonly understood in
terms of CL and no elaborate alternative understanding is available.
Although CL is of recent vintage, the historiography of the sciences did not as
yet change the outlook. Hundreds of historians of science have been working in the
field. Those that recurred to non-standard logics can still be counted on one's
fingers.
So, the situation is that scientists think to be using CL, that most historians think
scientists always (implicitly) did, and that logicians have no decent practical
alternative (nor results to prove the scientists and historians wrong). All this forces
one, at least for the time being, to concentrate on CL as the upper limit logic (and
normality standard).
The story about the lower limit logic is somewhat more complicated. Here I have
to tum to the relevantist and dialetheist programmes. The reason why I am not an
adherent of either should be clear. Both programmes aim at a single logical system
(possibly containing several implications and several variants of other logical
constants) that applies to all reasoning. If the views I defended in Section 3 are
correct, this aim cannot be realized.
Adherents of both programmes claim that their preferred logic contains CL.
Indeed, (i) the negation-disjunction-conjunction-fragment of E as well as LP is
identical to that of CL, and (ii) in CL, that fragment is functionally complete
(roughly: all formulas of the system are equivalent to formulas of that fragment).
The containment should be qualified by adding that, although all theorems of the
aforementioned fragment are theorems of E and LP (and similar logics), the
inferential force of the CL-formulas is not retained. Typically, Disjunctive
Syllogism (A V B, ~ A f- B) does not hold in those logics. Both relevantists and
dialetheists have argued that Disjunctive Syllogism is invalid anyway. So, the idea is
that the alternative systems have the same expressive power as CL, but that classical
logicians made a mistake when delineating the correct inferences-"what classical
22 That, in the finite models, all of the finitely many numbers are also different from themselves will not
convince most people that there are infinitely many numbers anyway.
140 DIDERIK BATENS
logicians really wanted (or should have wanted) to say, can be said in my preferred
system".
A second qualification is in order: the aforementioned logics contain the
theorems of CL, but at the expense of changing the meaning of the logical
constants. 23 If one considers consistent models only, ~A (the paraconsistent
negation of A) is true in the same models as -,A (the classical negation of A);
similarly, the implication A :J B is true in the same models as ~AV B. However,
precisely in inconsistent (and incomplete) models, ...,A and ~A are not true in the
same models, and similarly for A :J B and ~ A vB.
This second qualification is a subtle one. Relevantists and dialetheists disagree
that the meaning of the logical constants is changed. A well-documented discussion
of their arguments would require too many pages. I shall briefly (but fairly)
characterize those arguments, present my objections, and refer the interested reader
to the source texts.
Let me first phrase the qualification in terms of a classical metalanguage-one of
which CL is the underlying logic. The usual semantic clause for material
implication is: vM(A :J B) = I iff vM(A) = 0 or vM(B) = 1. In CL, vM(A) = 0 iff
VM( ~A) = 1; hence, vM(A :J B) = VM( ~AV B). In paraconsistent logics (including
relevant logics), vM(A) = 0 does not follow from VM( ~A) = 1 as
vM(A) = VM( ~A) = 1 is not excluded. Hence, A :J B does not mean the same thing
as ~AvB. And obviously the classical negation of A, -,A, means something
different from the paraconsistent negation of A, ~A.
Next, let us tum to the semantic systems presented by relevantists and dialethe-
ists for their logics. There are many semantic styles for relevant logics, some rather
complicated. For the present discussion, it is sufficient to restrict our attention to the
propositional level and to consider just one semantic style. The following one is easy
to understand. Let vM(A), the value of a wff A in model M, be one out of {t}, {f},
{t, J}, and 0, denoting respectively that A is true, false, both true and false, and
neither true nor false. 24 Negation is characterized by: t E vM(~A) iffJE vM(A), and
J E VM( ~A) iff t E vM(A). Conjunction and disjunction are characterized in a similar
way. It is not difficult to see that classical negation is easily defined, e.g., by
t E VM( -,A) iff t tI: vM(A); and J E VM( -,A) iff t E vM(A).25 Material implication is
simply defined by writing out the conditions for -,AV B.
If this were the end of the story, relevantists and dialetheists would be sorts of
bigots, forbidding people to use logical constants that are absolutely sensible on
their own understanding of logic. This is not the case. They think to have a reason
why, nevertheless, classical negation and material implication do not make sense.
Central (and somewhat similar) arguments are presented in Section 80.2 of
Anderson et al. 1992 and in Priest 1990. In both, it is pointed out that the above
23 From a formalist point of view, this follows immediately from the fact that the inferential force of the
logical constants is weakened.
24 Priest rules out the latter case. Relevantists retain it, as they want a negation that is not only
paraconsistent, but also paracomplete.
25 The intuitive idea is quite simple: if t E vM(A), A is true and hence vM(~A) = (fl. Otherwise,
vM(~A) = it}.
IN DEFENCE OF A PROGRAMME FOR HANDLING INCONSISTENCIES 141
semantic clause for classical negation contains a metalinguistic negation, and both
programmes require that 'the True Logic' should underlie the metalanguage. In
Anderson et al. 1992, the discussion is carried out in semantic terms, where (in my
present notation) t E vM(A) is read as "A is at least true", and f E vM(A) is read as
"A is at least false". In Priest 1990,~he same semantic expressions are read
respectively as "A is true" and "A is false", and the discussion is carried out in terms
of these. Central to both discussions is that the metalinguistic "not" should be
interpreted as a paraconsistent negation. According to Anderson et al. 1992, the
truth conditions for classical negation then reduce it to the paraconsistent negation.
According to Priest 1990, classical negation comes out as different from its classical
understanding-see also Priest 1999. 26
The argument is clearly sound and interesting. It is nevertheless baffling because
of the aspects of the issue that are left out of account. One. There is absolutely no
need, given the semantics, to use a negation in the definition of classical negation.
The following clause does just the same job: VM( -,A) = {t} iff vM(A) E { {f}, 0},
and vM(-,A) = {f} iff vM(A) E {{/},{/, f}}. So, the above argument to the effect
that classical negation reduces to paraconsistent negation breaks down. Two. In his
1990, Priest moves to the 'informal' semantics (A is true/false) to discuss the matter.
Relevantists might have followed his example. They already admitted (Anderson et
al. 1992, 497): "Relevance logicians have so far invariably used a classical
metalanguage [ ... ] "preaching to the heathen in his own language.",,27 They might
very well admit, in some near future, that even expressions like (in my terminology)
"t E vM(A)" are in the classical metalanguage. So, why do they not present a
semantics in their own metalanguage? The answer is obviously that they did not yet
develop, on the basis of their preferred logic, a theory that is sufficiently rich to
formulate a semantics (for example to handle functions from arbitrary sets to
arbitrary sets). Three. As I claimed that the relevantist and dialetheist programmes
are interesting, I certainly am in favour of giving both programmes the time to
develop. But wait a minute. For all we can imagine right now, a semantic system in
terms of the developed programmes will contain valuation functions mapping the set
of formulas to a set of values, and there will be a set of designated values and a set
of undesignated values. Suppose that {t} and {f} are still around, and that the first
is designated whereas the second is not. We then presumably will be able to express
that vM(-,A) = {t} iff vM(A) is un designated, and vM(-,A) = {f} iff vM(A) is
designated. If that is so, we will be able to define classical negation. Four. Once a
relevantist or dialetheist theory for handling functions will be around, things will get
serious. The question will not any more be whether some metalinguistic negation is
paraconsistent rather than classical, but whether t E {f} and hence t = f, which
simply comes to 0 = I if you play the game differently.
26 Both discussions contain further arguments against classical negation. The central type of argument
will be dealt with below in the text.
27 Graham Priest disagrees that former versions of the semantics of relevant logic or of his logic LP are
formulated in a classical metalanguage. In his view, the valuation functions have to be replaced by
relations (in order to avoid a semantic liar paradox), but that does not mean that the former
metalanguage was classical (personal communication).
142 DIDERIK BA TENS
The upshot is that, although both programmes deserve the time to be carried out,
their attacks on alternatives are somewhat unsubstantiated. It is very well possible
that the logic and mathematics of the past were largely mistaken. People will readily
admit this, but not before a viable alternative is available.
Even if we realize that the elaboration of both programmes requires time, there
are some urgent problems with their present claims. Consider the value 0 admitted
by relevantists. The only way to express that A receives this value is obviously that
"not t E vM(A) and notf E vM(A)". However, by the very reasoning of Anderson et
at. 1992 (p. 495), this reduces to "f E vM(A) and t E vM(A)", i.e. "vM(A) = {t,j}".
So, relevantists, "preaching to the heathen in his own language", refer to a truth
value that they cannot even express according to their present-day standards.
Dialetheism is affected by similar problems (see also Batens 1990). In his 1991,
Priest states that the consistency of a statement should be presupposed unless and
until proved otherwise. In Priest 1987 (§8.3), he argues that, although Disjunctive
Syllogism is invalid, "If a disjunction is rationally acceptable and one of the
disjuncts is rationally rejectable, then the other is rationally acceptable." Central to
the argument is that "joint rational acceptance and rejection are not possible"; as he
puts it on p. 128, they are "incompatible". So far so good. However, it is central for
Priest that "[t]he whole beauty of the dialetheic/paraconsistent solution to the
paradoxes is that the tired old distinction between object language and metalanguage
is finally put to rest" (Priest 1990, 202, but see also Priest 1987). So, the least we
might expect, is that he is able to express, in (what he considers as) the True Logic,
the central logical concepts he uses in his own writings on logic. However, there is
no way to express, in Priest's LP (even if it is extended with the modal implication
from Priest 1987) that r is consistent, or that A and B are 'incompatible' or 'not
jointly possible'. To be more precise, it is impossible to express all this in LP with
the required force. Consider the statement that A and B are incompatible. There are
only three ways to express this in the framework ofLP. The first is -(A&B). This is
equivalent to -A V - B, and is true in some models in which A&B is true. The
second (relying on the modal semantics of Priest 1987) is -O(A&B). This is
equivalent to D( -A V -B), and is true in some models in which o(A&B) is true. The
third is (A&B) - 1- or perhaps A - (B - 1-). But both of these are 'compatible'
with A&B-Priest constantly stresses that the trivial model is logically possible. 28
Allow me to make, with all due respect, a remark in diagnosis of the problem.
The only connective that, within the framework of CL, is able to cause trouble is
negation. If an inconsistency arises within some theory T, one should not throw
away T. More often than not, one should live with T for a while, and reason from it.
If there is a way to eliminate the inconsistency, it is usually found by reasoning from
T. From a contextual point of view, the simplest way to proceed is to use, e.g., CL
as the underlying logic, and, if an inconsistency arises, to move to a suitable
adaptive logic. This road is not open to relevantists and dialetheists (on their present
28 A further problem relates to Section 3. Anything can be rationally questioned in some context, and
fortunately so. And nothing is rationally acceptable in all contexts. So, either nothing is rationally
acceptable, or being rationally acceptable is not incompatible with being rationally rejectable.
IN DEFENCE OF A PROGRAMME FOR HANDLING INCONSISTENCIES 143
The stand taken by relevantists and dialetheists has its price. I have shown before
that material implication is definable (in the semantics presented by relevantists and
dialetheists) in such a way that it is detachable. One might reply to this that one does
not need material implication anyway. But consider ordinary set inclusion, as in "all
of this year's Ghent philosophy students wear spectacles". Even within a
paraconsistent framework, this statement is adequately expressed by means of a
material implication: (Vx)(Px :J Sx). But there is nothing relevant or necessary about
it. So, the material implication cannot be replaced by a relevant or strict one.
Precisely within a paraconsistent framework, the sentence cannot be rendered by
(Vx)(~PxVSx). For one thing, the latter may be true even if P ([. S. Moreover, the
set inclusion (and the reading (Vx)(Px:J Sx)) allows one to derive Sa from Pa,
whereas (in a paraconsistent context) (Vx)(~PxVSx) does not. To the best of my
knowledge, no dialetheist ever tackled this problem. Relevantists did, but a brief
look at their most recent proposal (Anderson et al. 1992, §75) is sufficient to
convince oneself that their approach (i) is far from elegant, (ii) is not proven to
29 This holds for relevantists, who in general believe that inconsistencies occur in theories but cannot be
true, as well as for dialetheists who believe that there arc true inconsistencies (even if consistency may
be presumed unless and until proven otherwise).
3D One of the many passages one may refer to in this connection is Anderson et al. 1992 (§80.2.4). Even
the opening paragraph is most significant: in "an environment of possibly inconsistent information"
(my italics) there is "no conceivable use" for classical negation.
31 Maybe this is exaggerated. Both dialetheists and relevantists have made attempts to justify reasoning
according to some other logics in terms of their own-see Anderson and Belnap 1959, Meyer 1973,
Anderson et al. 1992, §33, and Priest 1991.
144 DIDERIK BATENS
coincide with set inclusion, and (iii) will require quite some work to be integrated (if
it can) in a semantics that also characterizes their preferred relevant implications. 32
Whatever one's sympathy for the relevance programme and the dialetheic
programme, it seems to me that the outcome of the present section forces one to stay
close to CL, even with respect to the lower limit logic, in devising inconsistency-
adaptive logics that are meant to be applied to real life situations. This conclusion is
obviously a provisional one, but then we have to act now.
32 Remember that this semantics is formulated in the language of the heathen. Whether it even can be
phrased in the 'two-valued semantics' of Anderson et al. 1992 (§80.2) is not discussed.
33 Rescher and Brandom 1980 seems to take a new start for handling inconsistency, and unfortunately is
not employed-it could-to characterize the inference relations.
34 An inference relation is called flat if it does not depend on any non-logical criteria (preferences, etc.).
35 The "G" refers to Guido Vanackere, who, working on a different problem, first arrived at this type of
interpretation of a set of possibly inconsistent premises.
IN DEFENCE OF A PROGRAMME FOR HANDLING INCONSISTENCIES 145
36 That the dynamics is nevertheless real is illustrated by the fact that it can be semantically captured-
see, e.g., Batens 1995, 1998.
37 This is a kind of degenerative case, comparable to {p, q, r, s&-s}.
146 DIDERIK BA TENS
Even if the sources may be retraced, we often want to retain, in the final and
consistent theory, parts of the information provided by sources of which most
information is rejected. These parts, however, are never among the Strong
consequences. 38 In many cases, they are not even provided by the other consequence
relations. This is a result of the formu~ation dependence of the Rescher-Manor
Mechanism. Consider the set {p&q, (q :::) r)&~p}, and remark that r is not even a
Weak consequence of it. Nevertheless, we may very well want to have p, q, and r in
our final theory. The Rescher-Manor Mechanism does not allow us to do so (even if
we appeal to non-logical preferences).
Let me now try to offer a general view on the difference between the standard
adaptive approach and the Rescher-Manor Mechanism. The first leads to a rich set
of consequences. This set will make it clear where the inconsistencies reside and
how they are connected to each other (if they are). In view of these insights one may
invoke external (non-logical) preferences to weed out the inconsistencies. The
preferences need not to be assigned to the premises themselves. They may pertain to
consequences of the premises, including consequences of jointly inconsistent
premises. In all interesting cases, the preferences are not available beforehand. Only
a rich set of adaptive consequences will enable us to devise experiments for testing
some of them, or to judge the theoretical use and elegance of others. 39
The consequence relations from the Rescher-Manor Mechanism proceed in a
very different way. All but the Weak consequence relation 40 lead to a consistent set
of consequences even from inconsistent premises. But precisely this seems to be a
weakness. There is no reason to believe that an empirical (or even a mathematical)
inconsistent theory contains all the information required to devise its consistent
improvement. Nearly always, only external preferences will be able to settle that
matter. Rescher himself remarked this already in his 1964 (p. 37): "And while the
recognition of ambiguity does fall within the province of logic, its resolution is
inevitably an extralogical matter." So, it is in general a disadvantage that the
consequence relations from the Rescher-Manor Mechanism prejudge on the
resolution of the inconsistencies.
Of course, there are cases where some consequence relation from the Rescher-
Manor Mechanism delivers exactly what we want. In such cases, that consequence
relation is obviously adequate. The reasons for its adequacy are not of a logical
nature, but that holds for the adequacy of any logic in any specific context.
The second topic to be discussed in the present section is the richness of the
lower limit logic. We have seen that CLuN is an extremely poor logic. Many
paraconsistent logics are richer. CLuNs extends CLuN with both directions of
Double Negation, de Morgan properties, and all other rules that drive negations
inwards (to the level of primitive formulas). Priest's LP is the negation-conjunction-
disjunction fragment of CLuNs (Jacking a detachable implication). A richer
38 The Strong consequences are those CL-derivable from all maximal consistent subsets of the premises;
the Weak consequences those CL-derivable from some maximal consistent subsets.
39 Meheus 1993 contains further arguments to prefer adaptive logics to the Rescher-Manor Mechanism
for studying scientific problem solving with inconsistent constraints.
40 We have seen two paragraphs ago that even these do not always provide the consequences that we want
to retain in the consistent improvement.
IN DEFENCE OF A PROGRAMME FOR HANDLING INCONSISTENCIES 147
Up to now, the Ghent group has mainly worked on the adaptive logics themselves.
Given the novel character of the logics and the difficulties in getting them right and
clean, this is quite understandable. But, as a side-effect, there is only one well
documented case study from the history of the sciences (see Meheus 1993, 200+),
only a couple of preferential mechanisms are well-studied (see Vanackere 200+,
Verhoeven 200+, and Batens et al. 200+), only one and a half (standard) non-
monotonic logics have been reconstructed (see Batens 1994 and De Clercq 200+),
only some results are available about conceptual dynamics (see, e.g., Batens 1995).
An especially urgent problem, not for the adaptive logic programme, but for
everyone, is the road along which consistency is reached from inconsistent premises.
The matter is more complex than it appears at first sight. In Batens 1994, a
mechanism is described to produce consistent models out of inconsistent ones. It
works fine in combination with circumscription-one of the techniques to arrive at
(standard) non-monotonic logics. However, different preferential mechanisms might
require that the 'consistent choices' are defined in a different way. Only after
settling this, shall one be able to determine the use and effectiveness of different
lower limit logics.
A very important insufficiently studied topic concerns adaptive logics that are
not inconsistency-adaptive. In CL, all abnormalities surface as inconsistencies. This,
41 The price to be paid is that Addition and similar 'constructive' rules are invalid. They are restored by
the adaptive ANA whenever the inconsistencies of the premises permit so.
148 DIDERIK BA TENS
REFERENCES
Anderson, A. R. and N. D. Belnap, Ir. (1959), A Simple Treatment of Truth Functions. Journal of
Symbolic Logic 24, 301-302.
Anderson, A. R. and N. D. Belnap, Ir. (1975), Entailment. The Logic of Relevance and Necessity, volume
I. Princeton University Press.
Anderson, A. R., N. D. Belnap, Jr., and J. M. Dunn (1992), Entailment. The Logic of Relevance and
Necessity, volume 2. Princeton University Press.
Batens, D. (1980), Paraconsistent Extensional Propositional Logics. Logique et Analyse 90-91, 301-302.
Batens, D. (1985a), Dynamic Dialectical Logics as a Tool to Deal with and Partly Eliminate Unexpected
Inconsistencies. In The Logic of Discovery and the Logic of Discourse, J. Hintikka and F. Vandamme
(eds.), New York: Plenum Press, pp. 263-271.
Batens, D. (1985b), Meaning, Acceptance, and Dialectics. In Change and Progress in Modern Science, J.
C. Pitt (ed.), Dordrecht: Reidel, 1985, pp. 333-360.
Batens, D. (1986), Dialectical Dynamics within Formal Logics. Logique et Analyse 114,161-173.
Batens, D. (1989) Dynamic Dialectical Logics. In Paraconsistent Logic. Essays on the Inconsistent. G.
Priest, R. Routley, and J. Norman (eds.), Miinchen: Philosophia Verlag, pp. 187-217.
Batens, D. (1990), Against Global Paraconsistency. Studies in Soviet Thought 39, 209-229.
42 Most unpublished papers in the reference section (and many others) are available from the internet
address http://iogica.rug.ac.be/centrumlwritings.
IN DEFENCE OF A PROGRAMME FOR HANDLING INCONSISTENCIES 149
Batens, D. (I 992a), Do we Need a Hierarchical Model of Science? In Inference, Explanation, and Other
Frustrations. Essays in the Philosophy of Science, J. Earman (ed.), University of California Press,
pp.199-215.
Batens, D. (1992b), Menselijke kennis. Pleidooi voor een bruikbare rationaliteit. Leuven, Apeldoorn:
Garant.
Batens, D. (1994), Inconsistency-adaptive Logics and the Foundation of Non-monotonic Logics. Logique
et Analyse 145, 57-94, appeared 1996.
Batens, D. (l995), Blocks. The Clue to Dynamic Aspects of Logic. Logique et Analyse 150-152, 285-328,
appeared 1997.
Batens, D. (1998), Dynamic Semantics Applied to Inconsistency-adaptive logics. In Logical
Investigations, volume 5. Moscow: NAUKA, pp. 74-85.
Batens, D. (1999a) Inconsistency-adaptive Logics. In Logic at Work. Essays Dedicated to the Memory of
Helena Rasiowa, Ewa Orlowska (ed.), Heidelberg, New York: Physica Verlag (Springer),
pp.445-472.
Batens, D. (I 999b) Paraconsistency and its Relation to Worldviews. Foundations of Science 3,259-283.
Batens, D. (1999c), Zero Logic Adding up to Classical Logic. Logical Studies 2,15. (Electronic Journal:
http://www.logic.ru/LogStudl02/LS2.html ).
Batens, D. (2000), Rich Inconsistency-adaptive Logics. The Clash between Heuristic Efficiency and
a
Realistic Reconstruction. In Logique en perspective. Melanges ofJerts Paul Gochet. F. Beets and E.
Gillet (eds.), Brussels: Editions OUSIA, pp. 513-543.
Batens, D. (2000), A Survey of Inconsistency-adaptive Logics. In Batens et al. (2000), pp. 49-73.
Batens, D. (200+), Towards the Unification of Inconsistency Handling Mechanisms. Logic and Logical
Philosophy, in print.
Batens, D., J. Meheus, D. Provijn, and L. Verhoeven (200+), Some Adaptive Logics for Diagnosis. Logic
and Logical Philosophy, in print.
Batens, D., C. Mortensen, G. Priest, and J. P. Van Bendegem, eds. (2000), Frontiers of Para consistent
Logic. Baldock: Research Studies Press.
Benferhat, S., D. Dubois, and H. Prade (l997), Some Syntactic Approaches to the Handling of
Inconsistent Knowledge Bases: A Comparative Study. Part I: The Flat Case. Studia Logica 58,17-45.
Darden, L. (1991), Theory Change in Science: Strategies from Mendelian Genetics. Oxford University
Press.
De Clercq, K. (200+), Two New Strategies for Inconsistency-adaptive Logics. Logic and Logical
Philosophy, in print.
Earman, J. (1986), A Primer on Determinism. Dordrecht: Reidel.
Kuhn, T. S. (1962), The Structure of Scientific Revolutions. The University of Chicago Press, second
enlarged edition 1970.
Laudan, L. (1981), A Confutation of Convergent Realism. Philosophy of Science 48, 19-49.
Meheus, J. (1993), Adaptive Logic in Scientific Discovery: the Case of Clausius. Logique et Analyse
143-144,359-389, appeared 1996.
Meheus, J. (1997), Wetenschappelijke ontdekking en creativiteit. Een paging tot theorievorming op basis
van een conceptuele, methodologische en logische studie. Unpublished PhD thesis, Universiteit Gent
(Belgium).
Meheus, J. (2000), An Extremely Rich Paraconsistent Logic and the Adaptive Logic Based on It. In
Batens et al. (2000), pp. 189-201.
Meheus, J. (200+), Inconsistencies in Scientific Discovery. Clausius's Remarkable Derivation of
Carnol's Theorem. In Acta of the XXth International Congress of History of Science, G. Van Paemel
et al. (eds.), Brepols, in print.
Meyer, R. K. (1973), Intuitionism, Entailment, Negation. In Truth, Syntax and Modality. Hughues
Leblanc (ed.), Amsterdam: North-Holland, pp. 168-198.
Nersessian, N. (2002), Inconsistency, Generic Modeling, and Conceptual Change in Science. This
volume.
Norton, J. (1987), The Logical Inconsistency of the Old Quantum Theory of Black Body Radiation.
Philosophy of Science 54, 327-350.
Norton, J. (1993), A Paradox in Newtonian Gravitation Theory. PSA 1992,2,421-420.
Priest, G. (1987),ln Contradiction. A Study of the Transconsistent. Dordrecht: Nijhoff.
Priest, G. (1990), Boolean Negation and All That. Journal of Philosophical Logic 9, 201-215.
Priest, G. (1991), Minimally Inconsistent LP. Studia Logica 50, 321-331.
150 DlDERIK BATENS
Priest, G. (\999), What not? A Defence of Dialetheic Theory of Negation. In What is Negation?, D. M.
Gabbay and H. Wansing (eds.), Dordrecht: Kluwer, 1999, pp. 101-120.
Rescher, N. (I 964 ), Hypothetical Reasoning. Amsterdam: North-Holland.
Rescher, N. (\ 973), The Coherence Theory of Truth. Oxford: Clarendon.
Rescher, N. and R. Brandom (1980), The Logic of Inconsistency. A study in Non-Standard Possible-
World Semantics and Ontology. Oxford: Blackwell.
Rescher, N. and R. Manor (1970), On Inference from Inconsistent Premises. Theory and Decision 1,
179-217.
Smith, J. (1988), Inconsistency and Scientific Reasoning. Studies in History and Philosophy of Science
19,429-445.
Van Meirvenne, J. (1994), Indexed Inconsistency-adaptive Logic. Logique et Analyse 145, 41-55,
appeared 1996.
Vanackere, G. (\997), Ambiguity-adaptive Logic. Logique et Analyse 159, 261-280, appeared 1999.
Vanackere, G. (2000), HL2. An Inconsistency-adaptive and Inconsistency-resolving Logic for General
Statements that might have Exceptions. Journal of Applied Non-Classical Logics 10, 317-338.
Vanackere, G. (200+), Preferences as Inconsistency-resolvers: The Inconsistency-adaptive Logic PRL.
Logic and Logical Philosophy, in print.
Verhoeven, 1. (200+), Position Changes in Discussions. Logic and Logical Philosophy, in print.
Vermeir, T. (200+), Inconsistency-adaptive Arithmetic.To appear.
JOKE MEHEUS*
1. INTRODUCTION
This paper concerns problem solving in inconsistent contexts. It is usually taken for
granted that inconsistencies are false, and I shall not challenge this view here. 1
Resolving some inconsistency may constitute the very problem one tries to solve.
Alternatively, one may realize that one is (and for some time will be) unable to
resolve an inconsistency within some domain, but nevertheless aim at solving
another problem within that domain. In cases like this, one faces two difficulties.
The first is to distinguish between inferences that are sensible and those that are not.
The second is to determine when a solution to the problem is acceptable. For a
decision on the latter difficulty, mere derivability (by some appropriate logic) is not
sufficient. It should also be plausible that the solution will remain derivable after the
inconsistencies are resolved.
The history of the sciences exhibits several examples of problem solving
processes that involve inconsistencies. In nineteenth century thermodynamics, for
instance, Rudolf Clausius tried to derive the answer to the question whether Carnot's
theorem is valid from two incompatible approaches to thermodynamic phenomena
(see Meheus 1993 and 200a). Other examples are Planck's and Einstein's derivation
of Planck's law (see, for instance, Smith 1988) and Einstein's account of Brownian
motion (see, for instance, Nickles 1980). In each of these cases, reasoning from the
inconsistencies was seen as necessary to arrive at satisfactory problem solutions.
The reason is that the inconsistencies were relevant for the problems at issue, and
hence, could not be disregarded. Moreover, at the time the problems were attacked,
there were no sufficient grounds to resolve the inconsistencies in a particular way.
Hence, simply rejecting some of the findings in order to restore consistency would
have been an arbitrary decision.
The most popular approach to handling inconsistent theories or sets of beliefs
divides them up in consistent subsets. This approach is well-known from Rescher
and Manor 1970. A related but much more complex approach was originated by
Schotch and Jennings 1980 and has been defended with respect to historical cases by
Bryson Brown in his 1990. In all such cases, the idea is that some statements are
taken to follow from the inconsistent premises if and only if they are consequences
(according to Classical Logic-henceforth CL) of all consistent subsets of the
premises that fulfil certain conditions. The nature of these conditions may be
'logical' or preferential. Maximal consistt;nci is an example of the former sort, the
condition that the set contains specific well-established empirical statements is an
example of the latter sort.
Some authors have argued against even this rather soft paraconsistene handling
of inconsistencies. Both John Norton (1992) and Joel Smith (1988) have argued that
the actual handling of inconsistencies is not logic driven but content driven. They do
not deny that one reasons in inconsistent contexts. However, according to them, this
reasoning still proceeds in terms of CL. Whether a CL-consequence A of the
inconsistent premises, say f, is sensible (or acceptable) does not depend on the
question whether A is a CL-consequence of f itself-indeed, anything is a CL-
consequence of f -but on the question whether it is a CL-consequence of the
consistent but for the time being unknown improvement off.
At first sight, both consistent chunking and the content driven handling of
inconsistencies seem neat and clear. If one has reasons to believe that some part L\ of
an inconsistent set f will definitely belong to the consistent improvement of f, then
classical consequences of L\ may be taken as reliable steps in the problem solving
process. 4 These reasons may be that L\ is extremely well established or that it follows
from all consistent subsets off that fulfil certain conditions.
If one confronts these approaches with common insights on decidability and the
effectiveness of derivations, they tum out rather problematic. Let us first return to
consistent chunking. A minimally realistic approach forces us to consider
predicative sets of premises, and it is well-known that "is a consistent subset of' is
not a decidable predicate. Worse, there even is no positive test for it. 5 So, although
the statement that A is a CL-consequence of all maximally consistent subsets of f is
well-defined, it is by no means clear in which way one might establish this fact. It
follows that the consistent chunking approach might delineate, in a perfectly clear
way, the sensible or acceptable consequences of the inconsistent f, but that no
human has any use from this insight in order to handle f.
The content driven approach is even worse in this respect. It refers to a consistent
set of premises that is not a subset of the present premises. The consistent
improvement of f will, in all interesting cases, at best overlap with f. It will not and
cannot contain all off, but it will also contain statements that are not members off.
Because of this, the content driven approach is extremely troublesome from the
point of view of effectiveness. But there is more. Determining which 'parts' of an
2 A consistent subset l:!. off is maximally consistent ifand only if l:!.u {A} is inconsistent for any A E f -l:!..
3A logic is called paraconsistent ifand only ifit does not validate the Ex Fa/so Quodlibet, viz. A, -A r- B.
4 That is: reliable steps towards the articulation of the consistent improvement of the theory, or reliable steps
toward the solution of another problem within the presently inconsistent context.
5 Also "to follow from" is undecidable, but it has a positive test. The latter means that if one constructed a
proof of A from f, one may be sure that A follows from f, and if A follows from f, there is bound to
'exist' a proof of A from f, even if we may never find it. This does not obtain for "is a consistent subset
of'. Even if l:!. is a consistent subset off, there need not exist any finite construction that establishes this.
How TO REASON SENSIBLY YET NATURALLY FROM INCONSISTENCIES 153
logic driven view, all sensible consequences (those that follow according to some
paraconsistent system) are considered as acceptable. I shall argue that these
identifications are mistaken. When reasoning from inconsistencies, one cannot
demand that an inference should be acceptable in order to be sensible. Even
consequences that are unlikely to follo~ from the consistent improvement of the
theory may have an important heuristic value (in the sense that they lead to a better
understanding of the theory or that they help to decide how the inconsistencies
should be resolved), and hence, should be derivable from the inconsistent theory.
However, when dealing with the question whether some problem solution that is
derived from inconsistent premises should be accepted, its mere derivability (by
some appropriate logic) is not sufficient.
I shall proceed as follows. Starting from a specific example from the history of
the sciences, I first discuss some central characteristics of problem solving processes
that involve inconsistencies (section 2). This enables me to address the question
which requirements a logic should meet in order to account for this type of
reasoning process (section 3). I shall argue that we need a so-called inconsistency-
adaptive logic. These logics 'adapt' themselves to the specific inconsistencies
involved. Where an inconsistency is involved, the rules of inference of CL are
restricted in order to avoid triviality; everywhere else they behave exactly like CL. 8 I
shall also argue that we need a special kind of inconsistency-adaptive logic, namely
one that (from the point of view of laypersons) stays as close as possible to CL. In
section 4, I briefly present the inconsistency-adaptive logic ANA, and show that it
meets the requirements discussed in section 3. I end with some conclusions (section
5).
In this section, I briefly discuss the central steps in Clausius' derivation of Camot's
theorem. 9 This derivation formed an integral part of Clausius' attempt to reconcile
two incompatible approaches to the phenomena of heat and work. On the one hand,
there was the theory of Sadi Camot which stated that the production of work in a
heat engine results from the mere transfer of heat from a hot to a cold reservoir. On
the other hand, there was the view, especially advocated by James Prescott Joule,
that the production of work in a heat engine results from the conversion of heat into
work. If both approaches are combined, several contradictions follow, e.g., that the
production of work results from the mere transfer of heat and that it results from the
conversion of heat.
It is generally believed that the conflict between the two approaches could be
resolved by simply eliminating from Camot's theory the view that heat is conserved.
In line with this, Clausius' discovery is in general not considered as creative (see, for
8 The first inconsistency-adaptive logic was designed by Diderik Batens (see his 1986 and 1989).
Meanwhile, a whole variety of inconsistency-adaptive logics is available and the notion of an adaptive
logic has been generalized to include other types of 'logical abnormalities'-for an overview, see Batens
2000.
9 For more details on this specific example, I refer the reader to Meheus 1993 and 200+a.
How TO REASON SENSIBLY YET NA TURALL Y FROM INCONSISTENCIES 155
instance, Mach 1896, Clark 1976 and Psillos 1994). According to the common view,
Clausius only had to resolve the conflict between the two approaches (which was
quite trivial), and to combine the result.
I have argued elsewhere (Meheus 199 +b) that this view is mistaken. Resolving
the inconsistencies involved was far from trivial. It was only through a series of
inferences that Clausius was able to decide which findings had to be rejected or
modified. Moreover, as Clausius was working with an inconsistent set of premises
(unlike Camot), he even had to reconsider Camot's derivations.
One of Camot's results for which Clausius had to design a new derivation was
Camot's theorem. Interestingly, he designed two different proofs for this theorem.
Both are valid from the point of view of eL, and are moreover very similar to each
other (both are based on Reductio ad Absurdum from a hypothesis). Nevertheless,
Clausius considered only one of them as a valid derivation. In order to understand
Clausius' reasoning process, we should first have a look at Camot's proof.
Camot's theorem states that no heat engine is more efficient than a reversible
engine. By a reversible engine, Camot means an engine that (i) in the normal
direction, produces work and transfers heat from a hot to a cold reservoir, (ii) in the
reversed direction, consumes work and transfers heat from a cold to a hot reservoir,
and (iii) functions in such a way that both directions annul each other (if the same
amount of heat is transferred, the amount of work produced in the normal direction
equals the amount of work consumed in the reversed direction). In order to make the
discussion as transparent as possible, I shall use the term "super efficient engine" to
refer to an engine that, with a given amount of heat, produces more work than a
reversible engine.
Camot's proof is based on Reductio ad Absurdum from a hypothesis. In order to
proof that no engine is more efficient than a reversible engine, he supposes that the
contrary holds true, namely that there is an engine that is more efficient than a
reversible engine. Next, he shows that this hypothesis leads to a contradiction.
Finally, he rejects the hypothesis on the basis of this contradiction-or, which,
comes to the same, affirms its negation.
Three ideas are central for Camot's proof. First, there is the core principle of
Camot's theory, namely
PI the production of work by a heat engine results from the mere transfer
of heat from a hot to a cold reservoir.
10 Remember that the amount of work a reversible engine produces in the normal direction equals the
amount of work it consumes in the reversed direction. So, if it holds true that, with a given amount of
156 JOKEMEHEUS
w.
Q2 = Q3
Given all this, Carnot's proof is straightforward. Suppose that a super efficient
engine exists. In that case, it can be combined with a reversible engine as shown in
Figure I. This combined engine would produce work without the expenditure of
heat or any other change to itself or to its surroundings, and thus would form a
perpetual motion machine. Indeed, in view of PI, there can neither be an
expenditure of heat nor a change to the engines. It also follows from PI that Q, = Q2
and that Q3 = Q4. But then, in view of Q2 = Q3, it follows that Q, = Q4. Hence,
after a complete series of operations of the combined engine, both reservoirs return
to their original state. As perpetual motion machines are impossible, it follows that a
super efficient engine cannot exist.
We can now tum to Clausius. In his 1850, he presents a new derivation for
Carnot's theorem that is compatible not only with PI but also with the alternative
view that the production of work results from the consumption of heat. Central in
this proof is the idea that heat cannot flow by itself from a cold to a hot body. This
idea would later give rise to the notion of entropy. In a later text, Clausius reveals
heat, a super efficient produces more work than a reversible engine, it also holds true that, under the same
conditions, the former produces more work than the latter needs to extract an equivalent amount of heat
from a cold reservoir.
How TO REASON SENSIBLY YET NATURALL Y FROM INCONSISTENCIES 157
that he originally designed a proof that was much closer to Carnot's (see Clausius
1863, 313). Let us first have a look at this original proof.
The main difference between Carnot's proof and Clausius' original proof is that
the premises of the latter include not only the relevant parts of Carnot's theory, but
also a central principle from the alternative approach, namely
In this proof, Clausius starts from the hypothesis that a super efficient engine exist.
In that case, Clausius continues, it would be possible to combine it with a reversible
engine as shown in Figure I. In view of PI, such a combined engine would produce
work without the expenditure of heat or any other change to itself or to its
surroundings. However, in view of P2, this combined engine would consume an
amount of heat equivalent to the amount of work produced. Hence, it is not the case
that the combined engine would produce work without the expenditure of heat or
any other change to itself or to its surroundings.
At this stage, an inconsistency is derived. Hence, according to CL, Reductio ad
Absurdum can be applied in order to reject the hypothesis. However, as Clausius
himself remarks in his 1863, he did not consider this as a valid derivation of
Carnot's theorem. Surprisingly enough, the proof included in his 1850 (which he
accepted as valid) is also based on Reductio ad Absurdum from a hypothesis.
There are three differences between Clausius' final proof and Carnot's. The first
is that, like in Clausius' original proof, the premises include PI as well as P2. Next,
the arrangement of the two engines is different. In Carnot's proof the situation is
considered where only part of the work produced by the super efficient engine is
used to operate the reversible engine (see Figure 1). In Clausius' final proof the
situation is considered where all the work produced by the former is consumed by
the latter (see Figure 2). Note that in the latter arrangement, the amount of heat S
delivers to the cold reservoir, is smaller than the amount of heat R absorbs from it. II
The third difference concerns the principle used to refute the hypothesis. Carnot's
proof is based on the impossibility of a perpetual motion machine. Clausius' final
proof, however, is based on the principle that no engine can transfer heat from a cold
to a hot reservoir without the expenditure of work or any other change to itself or to
its surroundings.
Clausius' proof can now easily be represented. Suppose that a super efficient
engine exists. In that case, it can be combined with a reversible engine as shown in
Figure 2. This combined engine would transfer heat from a cold to a hot reservoir
without the expenditure of work or any other change to itself or to its surroundings.
Indeed, in view of P2, QI = Q2 + Wand Q3 + W = Q4. Hence, if Q2 < Q3, then
QI < Q4. As such an engine is impossible, it follows that there are no super efficient
engines.
II Remember that a super efficient engine, with a given amount of heat, produces more work than a
reversible engine. Hence, in order to produce the same amount of work as a reversible engine, it will
transfer a smaller amount of heat.
158 JOKEMEHEUS
Q2<Q3
As the reader will have noticed, Clausius' proofs are very similar to one another.
An important similarity is of course that both are based on Reductio ad Absurdum
from a hypothesis. Moreover, the premises of both proofs include Carnol's theory as
well as Joule's view. Against this background, the statements arrived at in the
second proof are 'just as true' as those arrived at in the first proof. So, why then did
Clausius not consider the original proof as valid?
It is important to note that CL does not enable us to answer this question. As I
mentioned before, both proofs are valid according to CL. Hence, rejecting one of
them as invalid is unjustified. However, also most paraconsistent logics are
inadequate to make sense of Clausius' reasoning process. The reason for this is that
most paraconsistent logics simply invalidate Reductio ad Absurdum (as well as
several other rules from CL). As a consequence, both proofs of Clausius are invalid.
Hence, accepting one of the proofs as valid is unjustified.
Some readers will conclude at this point that only a content driven approach can
account for this specific problem solving process. However, such an approach
cannot account for Clausius' reasoning. Among other things, a content driven
analysis presupposes that Clausius knew in advance which 'parts' of the inconsistent
set of statements would belong to the consistent reformulation. This condition was
not satisfied. Both the original proof and the final proof were constructed from the
(inconsistent) union of Carnot's theory and Joule's view before a decision was made
on which parts of this union had to be retained (see also Meheus 1993 and 200+a).
There is something more. A closer inspection of the two proofs reveals that there
is a logical difference between the two. As I explain in more detail in my 200+a, the
How TO REASON SENSIBLY YET NATURALLY FROM INCONSISTENCIES 159
12 I only assume here that all unproblematic consequences are included in the consistent improvement-see
also below.
160 JOKEMEHEUS
We have seen in the previous section that neither a content driven approach nor one of
the usual logic driven approaches can account for Clausius' reasoning. A content driven
account fails because Clausius was reasoning from the inconsistent premises, and not
from some consistent improvement for them. The usual logic driven approaches fail
because they cannot account for the fact that a specific type of argument was not in
general (in)validated.
What we seem to need in order to make sense of Clausius' reasoning is a logic that,
like a paraconsistent logic, invalidates Reductio ad Absurdum in the first proof, but that,
like CL, validates it in the second proof. The reason for invalidating Reductio ad
Absurdum in the first proof should be that the sentences involved in its application
behave inconsistently with respect to the premises. Similarly, the reason for validating
Reductio ad Absurdum in the second proof should be that here the sentences involved
behave consistently with respect to the premises. To many readers it may seem that
these peculiarities are far beyond logic. There is, however, one kind of logics, so-called
inconsistency-adaptive logics, that give us exactly what we need.
As I mentioned before, inconsistency-adaptive logics have the unusual property that
they localize the specific inconsistencies that follow from a given set of premises, and
adapt themselves to these. In practice, this means that some inference rules of CL are
turned into conditional rules: they can be applied provided that specific sentences
behave consistently. Thus, in an inconsistency-adaptive logic, Reductio ad Absurdum
from a hypothesis comes to the following: 13
(*) given a derivation of Band not-B from the hypothesis A, one may
conclude to not-A provided that B behaves consistently with respect to
the premises.
Note that (*) validates Reductio ad Absurdum in Clausius' final proof, but not in his
original proof. This is exactly what we need to understand his reasoning.
It is important to note that Clausius' derivation of Carnot's theorem is only one
example of a much more general mechanism. When dealing with inconsistent premises,
our reasoning often proceeds in a dynamical way. Consequences that are derived at
some stage may later be rejected. The reason for this may be that new premises are
added that cause additional inconsistencies. The reason may also be, as in the example
discussed above, that some statement is discovered to behave inconsistently with
respect to the premises.
Inconsistency-adaptive logics are the only kind of formal logics available today that
can make this kind of dynamics understandable. The reason for this is precisely that
they turn some rules of inference into conditional rules (instead of simply invalidating
them as is done in all other paraconsistent systems). I shall now argue, however, that
not every inconsistency-adaptive logic is adequate for this type of reasoning process.
There are at least three additional requirements such a logic should meet.
13 In some inconsistency-adaptive logics it is also necessary to require that the consistent behaviour of B is
not connected to the consistent behaviour of any other sentence-see Batens 1999.
How TO REASON SENSIBLY YET NATURALLY FROM INCONSISTENCIES 161
The first requirement is that the logic should allow for realistic reconstructions of
natural reasoning in inconsistent contexts. Most inconsistency-adaptive logics are
highly problematic in this respect. The reason is that all of them tum some of the most
natural rules of CL (like Disjunctive Syllogism and Modus Tollens) into conditional
rules. Moreover, rules that are intuitively, similar are treated in a different way. For
instance, in all inconsistency-adaptive logics that are based on CLuN (see Batens
1999), Modus Ponens is an unconditional rule, but Modus Tollens is a conditional rule.
As a consequence, these logics allow one to infer B from A :J B and A in the presence
of -A, but they do not allow one to infer -A from A :J Band -Bin the presence of B.
It is not realistic to presuppose that scientists make these distinctions when reasoning
from inconsistencies. Rather, we have every reason to believe that they implicitly use a
system that is as close as possible to CL.
The second requirement is that the logic should enable one to distinguish between
problematic and unproblematic consequences of an inconsistent set of premises. By an
unproblematic consequence, I mean one that is not related to the inconsistencies
involved. In the case of Clausius' final proof, for instance, the principle that heat cannot
flow by itself from a cold to a hot reservoir is not related to the inconsistency between
PI and P2 (whether this principle is retained or not has no influence on the conflict
between PI and P2).
An important adequacy criterion for consistent alternatives of an inconsistent set r
is that all unproblematic consequences of r should be derivable from them. This is
rational in view of the fact that, when replacing an inconsistent set of statements by a
consistent alternative, one wants to retain as much information as possible, and hence,
one does not want to eliminate more 'parts' of the original set than is necessary for
resolving the inconsistencies. In view of this, the fact that one is able to establish that
some consequence is unproblematic makes it plausible that it will remain derivable
from the consistent replacement. So, in some cases problem solutions derived from
inconsistent statements can be accepted on purely formal grounds: it suffices to
establish that they are unproblematic.
Evidently, logic cannot tell us whether some problematic consequence of an
inconsistent set should or should not belong to the consistent improvement. This
decision can only be taken on external grounds. However, a logical analysis may highly
reduce the number of external justifications needed.
The final requirement is that, for problematic consequences, the logic should be as
rich as possible. The reason for this is that problematic consequences may have an
important heuristic value. Clausius' final proofforms again a nice illustration. Suppose,
for instance, that Clausius did not immediately realize that an analogous proof can be
constructed that relies on PI. In that case, the proof would still provide a better insight
in the premises. Moreover, it could further the search for a proof that is independent
from the inconsistencies involved. Problematic consequences are heuristically
important for another reason as well. As a consistent alternative can never retain all
problematic consequences, the decision that a specific problematic consequence should
or should not follow from the consistent replacement at once allows one to eliminate
some alternatives. Let me give a simple propositional example to illustrate this.
Consider the set r = {p&q, - r, (p V s) J r, q J t}. One of the problematic
consequences of r is p J r. If one decides that p J r should be rejected, one can
162 JOKEMEHEUS
eliminate all alternatives from which p J r follows. If one decides that p J r should be
accepted, one can eliminate all alternatives from which both p and ~r follow. In either
case, one gains some information on how the consistent replacement should look like.
Note that something similar does not hold for unproblematic clauses off, for instance,
s J r. Neither the decision to accept this specific consequence of f nor the decision to
reject it provides information on how the conflict between p, ~r and p J r should be
resolved.
As may be clear from the previous section, inconsistency-adaptive logics differ from
each other with respect to which rules of CL they tum into conditional rules. In Meheus
2000, I presented the inconsistency-adaptive logic ANA, and showed that it validates
unconditionally all analysing rules of CL. Analysing rules are (roughly speaking) those
that enable one to reduce complex formulas to less complex ones: Disjunctive
Syllogism, Modus Tollens, Modus Ponens, Simplification, Double Negation, ... In
addition to this, ANA validates all constructive rules that are adjunctive (for instance,
A, B 1= A&B). The only rules of CL that are only conditionally validated are some
constructive rules that are weakening, for instance, Addition (A 1= A V B) and
Irrelevance (A 1= B J A), or that are 'paradoxical', for instance, A 1= AV(B&~B) and
A 1= A&(BV ~ B). In a proof format that allows for Reductio ad Absurdum from a
hypothesis (as in Camot's and Clausius' proofs), also this rule is conditional.
In view of these features, ANA leads to an extremely rich consequence set. What is
important for our present purposes, however, is that ANA allows for realistic
reconstructions of reasoning in inconsistent contexts. As the example of Clausius
illustrates, the restriction on Reductio ad Absurdum from a hypothesis is intuitively
justified. All other restrictions in ANA concern rules that are in general considered as
'unnatural' (like Addition and Irrelevance), and that are only used in very specific
circumstances. Hence, from the point of view of laypersons, ANA stays extremely
close to CL.
Another important feature of ANA is that its richness enables one to delineate, in a
quite natural way, the unproblematic consequences of an inconsistent set. Let P be the
set of all formulas that are obtained from f by first putting each member off in prenex
conjunctive normal form, and next applying Simplification (A&B 1= A; A&B 1= B) as
much as possible. The set of unproblematic 'clauses' off can be defined as n = {A I
A E P; f Ii=ANA ~ A}. In view of this definition, a statement may be considered as
unproblematic if and only if it is ANA-derivable from n.
14 What this intuitively comes
5. IN CONCLUSION
15 Analogously, pV q is unproblematic if and only if either p or q is. Hence, pV q is unproblematic if and only
if either -p or -q is not derivable or, which comes to the same, -(P&q) is not derivable.
16 Evidently, this only holds true for consistent alternatives that neither involve conceptual changes nor
incorporate new information.
164 JOKEMEHEUS
REFERENCES
Batens, D. (1986), Dialectical Dynamics within Fonnal Logics. Logique et Analyse 114,161-173.
Batens, D. (1989), Dynamic Dialectical Logics. In Para consistent Logic. Essays on the Inconsistent, G.
Priest, R. Routley and 1. Nonnan (eds.), Miinchen: Philosophica Verlag, 1989, pp. 187-217.
Batens, D. (1999) Inconsistency-adaptive Logics. In'Logic at Work. Essays Dedicated to the Memory oj
Helena Rasiowa, Ewa Orlowska (ed.), Heidelberg, New York: Physica Verlag (Springer), 1999,
pp. 445-472.
Batens, D. (2000), A Survey of Inconsistency-adaptive Logics. In Batens et al. (2000), pp. 49-73.
Batens, D., C. Mortensen, G. Priest, and J. P. Van Bendegem (eds.) (2000), Frontiers oj Paraconsistent
Logic. Baldock: Research Studies Press.
Brown, B. (1990), How to be Realistic about Inconsistency in Science. Studies in the History and
Philosophy oj Science 21, 281-294.
Clark, P. (1976), Atomism versus Thennodynamics. In Method and appraisal in the physical sciences.
The critical background to modern science. 1800-1905, C. Howson (ed.), Cambridge: Cambridge
University Press, pp. 41-105.
Clausius, R. (1850) Ueber die bewegende Kraft der Wanne und die Gesetze, welche sich daraus fUr die
Warmelehre selbst ableiten lassen. Reprinted in Ueber die bewegende Kraft der Wiirme und die
Gesetze, welche sich daraus for die Wiirmelehre selbst ableiten lassen, M. Planck (ed.), Leipzig:
Verlag von Wilhelm Engelmann, 1898, pp. I-52.
Clausius, R. (1863) Ueber einen Grundsatz der mechanischen Warmetheorie. Reprinted and translated in
Theorie mecanique de la chaleur par R. Clausius. F. Folie (ed.), Paris: Librairie scientifique,
industrielle et agricole, 1868, pp. 311-335.
Mach, E. (1896), Die Principien der Wiirmelehre. Leipzig: Verlag von Johann Ambrosius Barth.
Meheus, J. (1993), Adaptive Logic in Scientific Discovery: The Case of Clausius. Logique et Analyse
143-144,359-391, appeared 1996.
Meheus, J. (1999), Clausius' Discovery of the First Two Laws of Thennodynamics. A Paradigm of
Reasoning from Inconsistencies. Philosophica 63, 89-117.
Meheus, J. (2000), An Extremely Rich Paraconsistent Logic and the Adaptive Logic based on It. In
Batens et al. (2000), pp. 189-20 I.
Meheus, J. (200+a), Inconsistencies in Scientific Discovery. Clausius' Remarkable Derivation of
Carnot's Theorem. In Acta oJthe XXth International Congress ojHistory oJScience, G. Van Paemel
et al. (eds.), Brepols, in print.
Meheus, J. (200+b), On the Acceptance of Problem Solutions Derived from Inconsistent Constraints.
Logic and Logical Philosophy, in print.
Nickles, T. (1980), Can Scientific Constraints be violated Rationally? In Scientific Discovery, Logic. and
Rationality. T. Nickles (ed.), Dordrecht: Reidel, 1980, pp. 285-315.
Norton, J. D. (1992), A Paradox in Newtonian Cosmology. PSA 1992, vol. 2, pp. 412-420.
Psillos, S. (1994), A Philosophical Study of the Transition from the Caloric Theory of Heat to
Thennodynamics: Resisting the Pessimistic Meta-Induction. Studies in the History and Philosophy oj
Science 25, 159-190.
Rescher, N. and R. Manor (1970) On Inference from Inconsistent Premises. Theory and Decision I,
179-217.
Schotch, P. K. and R. E. Jennings (1980), Inference and Necessity. Journal oj Philosophical Logic 9,
327-340.
Smith, J. (1988), Inconsistency and Scientific Reasoning Studies in History and Philosophy oJScience 19,
429-445.
ERIK WEBER AND KRISTOF DE CLERCQ*
1. INTRODUCTION
1.1 In the philosophical literature, the natural phenomena for which science can
provide explanations are usually divided into particular events and general laws.
For instance, Wesley Salmon writes:
Scientific explanations can be given for such particular occurrences as the appearance of
Hailey'S comet in 1759 or the crash of a DC-I 0 jet airliner in Chicago in 1979, as well
as such general features of the world as the nearly elliptical orbits of planets or the
electrical conductivity of copper. [1984, 3]
* Kristof De Clercq is a Research Assistant of the Fund for Scientific Research - Flanders.
165
J. Meheus (ed.), Inconsistency in Science, 165-184.
© 2002 Kluwer Academic Publishers.
166 ERIK WEBER AND KRISTOF DE CLERCQ
Kitcher ascribes to Hempel the view that, besides expectability, explanations can
confer a second intellectual benefit upon us: unification. Whether or not this
ascription is correct does not matter: the important thing is that Kitcher regards
unification as the one and only benefit that explanations may produce. A third
influential position is Wesley Salmon's:. in his 1984, he claims that knowing the
causal mechanisms that produce the events we observe, is the intellectual benefit
explanations can confer upon us.
In our view, there are several intellectual benefits explanations can confer upon
us. We think that, besides insight in causal mechanisms, unification and
expectability, there is at least one more possible benefit: explanations can give
meaning to the events they explain. To avoid misunderstanding, it must be stressed
that we do not claim that all explanations give all these benefits. There are different
types of explanations of events: nomological explanations (which aim at
expectability), unificatory explanations, causal explanations, etc. Each type has its
characteristic intellectual benefits. A similar pluralistic view on explanation and
understanding is defended in Salmon 1993 and 1998.
1.2 Hempel used Classical Logic (CL) to define nomological explanations. His
formal elaboration of the nomic expectability idea can be summarized as follows:
The first aim of this article is to show that inconsistency-adaptive logics are much
better suited than CL for defining nomological explanations. Inconsistency-adaptive
logics 'oscillate' between a paraconsistent lower limit logic (i.e. a logic which does
not validate Ex Falso Quodlibet) and an upper limit logic, usually CL. In section 2
we show that a CL-based approach to nomological explanations runs into serious
problems; in section 3 we briefly discuss the inconsistency-adaptive logic ACLuNl,
and present a formulation for it that is especially suited for the kind of application
we have in mind. In section 4 we show how the problems discovered in 2 can be
solved by means of this logic.
The second aim of this paper is to show that our conclusion with respect to
nomological explanations can be extended to other types of explanation of particular
events. In section 5 we argue that we also need an inconsistency-adaptive logic in
order to explicate how unification is reached. Section 6 deals with causal
understanding. Some people may doubt whether logic matters for this kind of
understanding. We argue that it does and that CL is insufficient.
In the last section we discuss explanation of laws. Again, we argue that CL is not
a good tool for explicating this kind of explanation.
WHY THE LOGIC OF EXPLANATION IS INCONSISTENCY -ADAPTIVE 167
E: Nixon is a pacifist.
This example shows that there are inconsistent knowledge systems, like Kj, for
which there is an E and -E such that both are explained (in Hempel's sense) while
no contrast is established: both E and - E can be expected, since both are
deductively derivable.
2.3 The most obvious way to cope with the problem established in 2.2, is to add a
clause to Hempel's definition:
Suppose we observe that Tweety does not fly. This fact cannot be explained,
because we can derive both that Tweety flies and that he does not fly. Thus far there
WHY THE LOGIC OF EXPLANATION IS INCONSISTENCY -ADAPTIVE 169
is no problem. But suppose that we observe that Billy flies. We might propose the
following argument as explanation:
L: Birds fly.
C: Billy is a bird.
E: Billy flies.
This argument satisfies the conditions (i)-(iv) of N2; however, this definition
classifies it as an inadequate explanation because the negation of the explanandum
can be derived from the knowledge base. Indeed, because of Ex Falso Quodlibet we
can derive from K2 that Billy does not fly: as soon as we have derived the
contradiction that Tweety flies and does not fly, every other statement can be
derived.
The example illustrates the general problem with N2 : though the information we
have about some facts (e.g. that Billy flies) is consistent, the inconsistency of the
knowledge system as a whole implies that nothing can be explained. In section 4 we
will present a definition which guarantees that an explanation always establishes an
incompatible contrast, but unlike N2 does not imply that all inconsistent knowledge
systems are explanatorily worthless. For instance, our definition will entail that K2
nomologically explains that Billy flies, but not that Tweety flies or does not fly. The
crucial concept in our definition will be ACLuNl-derivability.
1 In Batens et al. 200+ it is shown that our Fitch-style presentation as given below is equivalent to the
original formulation of ACLuNl in terms of somewhat complex generic rules (as in Batens 1998).
170 ERIK WEBER AND KRISTOF DE CLERCQ
3.1 ACLuNl-proofs are written in a specific format according to which each line of
a proof consists of five elements:
(i) a line number,
(ii) the well-formed formula (henceforth: wff) derived,
(iii) the line numbers of the wffs from which (ii) is derived,
(iv) the rule of inference that justifies the derivation, and
(v) the formulas on the consistent behaviour of which we rely in order for (ii) to be
derivable by (iv)from the formulas of the lines enumerated in (iii).
The fifth element causes the dynamics of ACLuNl-proofs: a line will be "deleted"
whenever the condition expressed in its fifth element is not (any longer) fulfilled.
In order to give a Fitch-style presentation of ACLuNl, we first need some
definitions from Batens 1998. Where A&~A is a formula in which the variables Ut,
... , Uk (k ~ 0) occur free (in that order), let ::l(A&~A) be (::lUI) ... (::luk)(A&~A). Let
DEK{A 1, ••• , An} refer to ::l(AI&~AI)V ... V::l(An&~An)' a disjunction of (where
necessary) existentially quantified contradictions.
Definition 1.
A formula A occurs unconditionally at some line of a proof iff the fifth element
of that line is empty.
Definition 2.
A behaves consistently at a stage of a proof iff ::l(A&~A) does not occur
unconditionally in the proof at that stage.
Definition 3.
The consistent behaviour of A 1 is connected to the consistent behaviour of A 2, ••• ,
An at a stage of a proof iff DEK{A t, ... , An} occurs unconditionally in the proof at
that stage whereas DEK(tJ.) does not occur unconditionally in it for any tJ. C {A t,
... , An}.
Definition 4.
A is reliable at a stage of a proof iff A behaves consistently at that stage and its
consistent behaviour is not connected to the consistent behaviour of other
formulas.
2 The index t;. in e.g. A :::) Bd obviously refers to the entire formula A :::) B, indicating that the formula
A :::) B occurs at a line the fifth element of which is t;..
WHY THE LOGIC OF EXPLANATION IS INCONSISTENCY -ADAPTIVE 171
f.., and (ii) relettering the variables in each member of f.. to the effect that they all
occur alphabetically.
STRUCTURAL RULES
Premise rule
PREM At any stage of a proof one may add a line consisting of (i) an
appropriate line number, (ii) a premise, (iii) a dash, (iv) 'PREM', (v)
'0'.
Hypothesis rule
HYP At any stage of a proof one may add a line consisting of (i) an
appropriate line number, (ii) an arbitrary wff, (iii) a dash, (iv) 'HYP',
(v) '0'. This line starts a new subproof.
Reiteration rule
REIT In a subproof one may add a line consisting of (i) an appropriate line
number, (ii) a wff that occurs in the main proof or in a non-terminated
subproof, (iii) the number of the reiterated line, (iv) 'REIT' , (v) the
fifth element of the reiterated line.
INFERENCE RULES
The rules of inference of ACLuNl are of two kinds: unconditional rules and
conditional rules. It is permitted to apply any of these rules at any stage of the proof,
even if applying them leads to an inconsistency. As usual, u and ~ should be
interpreted in such a way that all (main) formulas are wffs.
Unconditional rules
MP A :J B/1, A0 / B/1u0
CP to derive A :J BtJ, from a proof of BtJ, on the hypothesis Ae
PEIRCE (A :J B) :J A/1 / A/1
CON] A/1, B0 / A&B/1u0
SIM A&B/1 / A/1
A&B/1 / BtJ,
ADD AtJ,/AVB/1
B/1 / AV B/1
DIL AV B/1, A :J C0 , B :J C:::: / C/1u0u::::
IE A :J B/1, B :J A0 / A == BtJ,u0
EE A == B/1 / A :J B/1
A == B/1 / B :J A/1
EM A :J B/1, -A :J B0 / B/1u0
UI (Vu)A(u) /1 / A(~h
VO A(~)/1 / (Vu)(A(u))tJ,(alP) provided that ~ does not occur in a premise, in
a non-eliminated hypothesis, or in A(u).
172 ERIK WEBER AND KRISTOF DE CLERCQ
EG A(~),.. / C3a)A(a),..
MPE (3a)A(a) ,.., A(~) ::> Be / B"'ve(alp) provided that ~ does not occur in a
premise, in a non-eliminated hypothesis, in A(a), or in B.
II ... /a=a0 ,
EI a = ~ ,.., Ae / B,..ve where B is the result of replacing in A an
occurrence of a outside the scope of a negation by ~.
Conditional rule:
CR AV3(B&-B),.. / A"'v{B}
The difference between unconditional rules and the conditional rule is that, by
applying the latter, new members are introduced in the fifth element of the line
derived.
MARKING RULE
At any stage of the proof, it is obligatory to apply the following marking rule:
MR If A is not (any more) reliable, then "delete" from the proof all lines
the fifth element of which contains A (actually, those lines will be
marked, indicating that they are OUT).
To speed up proofs, we give some interesting conditional rules which can be derived
within ACLuNl:
Let us analyse a very simple ACLuNl-proof, that will clarify the specific
characteristics of the inconsistency-adaptive logic ACLuNl.
WHY THE LOGIC OF EXPLANATION IS INCONSISTENCY-ADAPTIVE 173
1 pVq PREM 0
2 -p PREM 0
3 -rVs PREM 0
4 r&-q PREM 0
5 t -:::J -q PREM 0
6 q 1,2 CDS {p} OUT at stage 14
7 -t 5,6 CMT {p, q} OUT at stage 14
8 r 4 SIM 0
9 -q 4 SIM 0
10 p 1,9 CDS {q} OUT at stage 14
11 (PVq)&~p 1,2 CONJ 0
12 (P&~p)Vq 11 DIST3 0
13 ((P&~p)Vq)&~q 9,12 CONJ 0
14 (P&~p)V(q&~q) 13 DIST 0
15 s 3,8 CDS {r}
3 DlST is an unconditional rule that is derivable within ACLuNl: it stands for (AvB)&C" / (A&C)VB"
and its variants like (A V B)&C" / A V (B&C)" (see line 14).
174 ERIK WEBER AND KRISTOF DE CLERCQ
Definition 5
A is finally derived at some line in an ACLuNl-proof iff (i) A is the second
element of the line and (ii) where tl is the fifth element of the line, any extension
of the proof can be further extended in such a way that it contains a line that has
A as its second element and tl as its fifth element.
Obviously, each formula that occurs at a line the fifth element of which is empty, is
finally derived (as that line cannot possibly be marked in any extension of the
proof).
Definition 6
r I- ACLuNI A (A is an ACLuNl-consequence of r or is finally derivable from f)
iff A is finally derived at some line in an ACLuNl-prooffrom r.
3.3 In CL, it is the combination of specific weakening rules (as Addition and
Irrelevance) and specific analysing rules (as Disjunctive Syllogism and Modus
Tollens) that leads to Ex Falso Quodlibet (EFQ). SO in order to avoid EFQ, and by
this obtaining a (non-adaptive) paraconsistent logic 4 , one has to drop either some
analysing rules or some weakening rules of CL. As almostS all paraconsistent logics
drop Disjunctive Syllogism and Modus Tollens, we will briefly focus on one such
logic, the (very weak) paraconsistent logic CLuN from Batens 1998.
A Fitch-style presentation of CLuN is easily obtained: it consists merely of the
structural rules and unconditional rules of ACLuNl, hence CLuN has neither
conditional rules nor a marking rule. 6 CLuN is a very poor paraconsistent logic: it
comes to full positive logic to which the axiom A v ~ A is added. One can easily
construct richer extensions by adding some or all of the following rules (and/or their
converses ):
DN ~~AIA
NI ~(A J B) I (A&~B)
NC ~(A&B)I ~Av~B
ND ~(AVB)IA&~B
NV ~(Va)A I (:3a)~A
4 Syntactically, paraconsistent logics can be characterized as logics that do not validate Ex Falso
Quodlibet.
S An interesting exception is Meheus' paraconsistent logic AN from Meheus 2000, which validates
Disjunctive Syllogism and Modus Tallens, while other rules of CL (Addition, Irrelevance and certain
forms of distribution) are invalid in it.
6 As the fifth element is always the empty set, all references to this fifth element may be omitted.
WHY THE LOGIC OF EXPLANATION IS INCONSISTENCY-ADAPTIVE 175
N3 ~(3a)A ! (Va)~A
Let PL be any thus obtained extension of CLuN. In the next section we will show
that for our purposes an inconsistency-adaptive logic is to be preferred above PL. In
a general way, the advantage of using ACLuNl can be put as follows. PL restricts
at least some inference rules of CL globally: e.g. Disjunctive Syllogism and Modus
Tollens are incorrect in general. As a result, these rules cannot even be applied to
consistent parts of the knowledge system: but if C and -Cv D are derivable from the
premises, and C does not behave inconsistently on those premises, then why should
we not derive D? In contradistinction with CLuN and its extensions, which classify
some CL-rules as incorrect, ACLuNl classifies only certain applications of CL-
rules as incorrect with respect to the set of premises. This implies that, with respect
to consistent parts of the knowledge system, ACLuNl preserves the full force of
CL, whereas PL has weaker inferential force. For similar reasons ACLuNl is to be
preferred to all other non-adaptive paraconsistent logics, as at least one CL-rule will
be globally invalid in the latter.
4.1 We first define neg(A): if A is of the form ~B, then neg(A) = B; otherwise
neg(A)= ~A. The alternative definition of nomological explanation we propose is:
(N3) Knowledge system K provides a nomological explanation for E if and
only if
(a) there are singular sentences C 1, ... , Cn and a law L such that
(i) {CI, C2, ... , Cn, L} is consistent,
(ii) C 1&C2&",&Cn and L are ACLuNl-derivable from K,
(iii) E is ACLuNl-derivable from C 1&C2& ... &Cn&L,
(iv) E is ACLuNl-derivable from K, and
(v) E is not ACLuNl-derivable from C 1&C2& ... &Cn ;
(b) neg(£) is not ACLuNl-derivable from K.
4.2 Since {C[, C2, ... , Cm L} is consistent, clauses (iii) and (v) ofN3 are equivalent to
the corresponding clauses in definition N 2. To see what difference clauses (ii) and
(b) make, let us derive some statements from K2 by means of ACLuNl: 7
1 (Vx)(Bx ~ Fx) PREM 0
2 (Vx)(Px ~ ~Fx) PREM 0
3 Bt&Pt PREM 0
4 Bb&~Pb PREM 0
5 Bt ~ Ft 1 UI 0
6 Bb ~ Fb 1 UI 0
7 Pt ~-F 2 UI 0
8 Pb ~ -Fb 2 UI 0
9 Bt 3 SIM 0
10 Pt 3 SIM 0
11 Bb 4 SIM 0
12 -Pb 4 SIM 0
13 Ft 5,9 MP 0
14 Fb 6,11 MP 0
15 -Ft 7,10 MP 0
The main difference between this definition and (N3) is illustrated by the following
example:
"Tweety is not a tiger" can be derived from K3 by means of ACLuN1, but not by
means of CLuN:
Since Modus Tollens is not valid in CLuN, -Tt cannot be derived in this logic. As a
consequence, it is inadequate for defining nomological explanations. Analogous
examples could be given in which CDS is used instead of CMT. Since Modus
Tollens and Disjunctive Syllogism are also missing in the stronger logics that can be
obtained by adding to CLuN some of the rules mentioned in 3.3, this whole class of
logics is inadequate for our purposes. As we mentioned above, some non-adaptive
paraconsistent logics validate Disjunctive Syllogism as well as Modus Tollens.
However, also in these logics it holds true that some CL-rules are incorrect in
general. Therefore analogous counterexamples can be constructed for these logics.
5. UNIFICATION
L L
Cal &Ca2 &,,,&Can Cbl&Cb2&",&Cbn
The crucial feature is that the two explananda, Ea and Eb , are explained by means of
the same law L but with different initial conditions. This is illustrated in the
following example:
The phenotypes of the ABO blood group system (the blood groups A, B, AB and 0)
are determined by the genes lA, IB and 10. IArAxIAro is a category of cross: an
rArAxrArO -individual is a descendant from one parent with genotype rAIA and one
parent with genotype IAlo. One can construct a number of analogous explanations, in
which the occurrence of blood group A is' explained for other people than Mary, but
the same law L is used. So there is a whole series of facts that we can show to be
instances of the law "All humans which belong to category IAIAxIAro have blood
group A".
5.2 The best possible CL-based definition of unification IS one that takes into
account Van Fraassen's critique on Hempel:
This definition has the same drawback as (N 2): it deprives inconsistent knowledge
systems from all explanatory power. By substituting "CL-derivable" in U 1 for
"ACLuNl-derivable" and adding two clauses similar to clause (iv) in definition (N3)
we obtain an adequate definition. That non-adaptive paraconsistent logics result in
inadequate definitions can be shown by examples analogous to the one given in
section 4.2.
6. CAUSAL UNDERSTANDING
6.1 Causal explanations describe the causal mechanisms that produced an event that
we have observed. As an example, consider two objects 01 and 02. We observe that
WHY THE LOGIC OF EXPLANATION IS INCONSISTENCY-ADAPTIVE 179
02 at t has a velocity of 2/3 (we consider only motion in one direction). The
following explanation may be given:
At t' (t' < t) there was a collision between 01 and 02. This collision was a causal
interaction in which the velocity of Ol'was changed from 1 into -1/3, and that of
02 from 0 into 2/3. In the period between t' and t the velocity of 2/3 was
spontaneously preserved by 02.
There are various kinds of causal explanations. The simplest ones contain only two
elements: a description of a causal interaction and a claim about spontaneous
preservation of a property. The concept of causal interaction was introduced by
Wesley Salmon (1984) in order to cover the innovative aspect of causation. There
has been a lot of discussion about what is the best way to define causal interactions
(see Dowe 1992, Salmon 1994, Dowe 1995). This discussion is not relevant for our
purposes. We will adopt a definition that is very close to Salmon's original
definition:
An object can be anything in the ontology of science (e.g. atoms, photons, ... ) or
common sense (humans, chairs, trees, ... ). Collision is the prototype of causal
interaction: the momentum of each object is changed, this change would not have
occurred without the collision, and the new momentum is preserved in an interval
immediately after the collision. When a white light pulse goes through a piece of red
glass, this intersection is also a causal interaction: the light pulse becomes and
remains red, while the filter undergoes an increase in energy because it absorbs
some of the light. The glass retains some of the energy for some time beyond the
actual moment of interaction. As an example of an intersection which is not a causal
interaction, we consider two spots of light, one red and the other green, that are
projected on a white screen. The red spot moves diagonally across the screen from
the lower left-hand comer to the upper right-hand comer, while the green spot
moves from the lower right-hand comer to the upper left-hand comer. The spots
meet momentarily at the centre of the screen. At that moment, a yellow spot appears,
but each spot resumes its former colour as soon as it leaves the region of
180 ERIK WEBER AND KRISTOF DE CLERCQ
1 VI =1 PREM
2 V2 = 0 PREM
3 ml = 1 PREM
4 m2 = 2 PREM
5 mivi + m2v2 = mlv{ + m2v{ PREM
6 v{ - v{ = - E( VI - V2) PREM
7 E=1 PREM
8 v{ - v{ = V2 - VI 6, 7
9 mlvl' + m2v{ = 1 1-5
10 v{ + 2v{ = 1 3, 4, 9
11 v{ - v{ = -1 1, 2, 8
12 VI' = v{- 1 11
13 (v{- 1) + 2v{= 1 10, 12
14 3v{ = 2 l3
15 v{ = 2/3 14
7. EXPLANATION OF LAWS
7.1 Consider three observable variables (C, D and E) that can have two values (0
and 1). Suppose that the following empirical laws have been found:
L I : (Vx)(Cox J EoX)
L 2 : (Vx)(Clx&DoX J Elx)
L3: (Vx)(Clx&Dlx J EoX)
Suppose further that we devise the following theory to explain the empirical laws:
182 ERIK WEBER AND KRISTOF DE CLERCQ
Theoretical laws
(Vx)(AoX J BoX)
(Vx)(A\x J B\x)
(VX)(A2X J BoX)
Auxiliary hypotheses
(Vx)(CoX J AoX)
(Vx)(BoX J EoX)
The theoretical laws relate the unobservable variables A and B. As can be derived
from the laws, we assume that B has two possible values, while A has three possible
values. The auxiliary hypotheses relate the unobservable variables A and B with
(some of) the variables featuring in the empirical laws.
T\ explains L\ in the following sense:
L2 and L3 are not explained by Tt. If we adopt definition E\, this problem can be
solved in "normal" but also counterintuitive ways. A normal way would be to add
the following auxiliary hypotheses to T\:
(Vx)(B\x J E\x)
(Vx)(C\x&DoX J A\x)
(Vx)(C\x&D\x J A2X)
Tz, i.e. the theory that results from adding these to hypotheses to T\, explains the
three empirical laws.
A counterintuitive procedure would be to add the following hypotheses to T\:
(Vx)(B\x J E\x)
(Vx)(C\x&D\x J A\x)
(Vx)(C\x&D\x J A2X)
(3x)(C\x&D\x)
7.2 A definition that blocks the counterintuitive moves can be easily formulated
within Classical Logic:
WHY THE LOGIC OF EXPLANATION IS INCONSISTENCY -ADAPTIVE 183
This definition has the same drawback as similar definitions discussed in the
previous sections: it eliminates too much. Indeed, E2 entails that inconsistent
theories do not explain any laws.
As can be expected, ACLuNl provides a way out:
With respect to T, and T2, this definition leads to the same judgments as E, and E2:
T, explains only L" while T2 explains L,-L3. The judgments about T3 differ
radically. As we already said, E, implies that T3 explains everything, while E2
implies that it explains nothing. Definition E3 implies that T3 explains L" but not L2
or L3:
(a) (Vx)(CoX::) EoX) is ACLuNl-derivable from T3 , but (Vx)(CoX&-EoX) is not
ACLuNl-derivable (the only existential assumption in T3 is that class C, is not
empty; Co can be empty) and neither is -(Vx)( CoX ::) EoX); so the first empirical
law is explained by T3
(b) (Vx)(C,x&DoX ::) E,x) is not ACLuNl-derivable from T3
(c) (Vx)(C,x & D,x ::) EoX) is ACLuNl-derivable from T3 , but so is
(Vx)(C,x&D,x&-EoX).
8. CONCLUDING REMARKS
8 The condition that both -(Vx)(Px ::l Qx) and (Vx)(Px&-Qx) may not be ACLuNl-derivable from T, is
necessary because they are not unconditionally equivalent in ACLuNl.
184 ERIK WEBER AND KRISTOF DE CLERCQ
REFERENCES
Batens, D. (1989), Dynamic Dialectical Logics. In Paraconsistent Logic. Essays on the Inconsistent. G.
Priest, R. Routley and J. Norman (eds.), Miinchen: Philosophia Verlag, pp. 187-217.
Batens, D. (1995), Blocks. The Clue to Dynamic Aspects of Logic. Logique et Analyse 150-152, 285-328.
Batcns, D. (1996), Functioning and Teachings of Adaptive Logics. In Logic and Argumentation, J. Van
Benthem, F. H. Van Eemeren, R. Grootendorst, F. Veltman (eds.), Amsterdam: North-Holland,
pp. 241-254.
Batens, D. (1998), Inconsistency-adaptive Logics. In Logic at Work. Essays Dedicated to the Memory of
Helena Rasiowa, E. Orlowska (ed.), Heidelberg, New York: Springer, pp. 445-472.
Batens, D. (2000), A Survey of Inconsistency-adaptive Logics. In Frontiers of Para consistent Logic, D.
Batens, C. Mortensen, G. Priest and J. P. Van Bendegem (eds.), Baldock: Research Studies Press,
2000, pp. 79-73.
Batens, D., K. De Clercq and G. Vanackere G. (200+), Simplified Dynamic Proof Formats for Adaptive
Logics. To appear.
Batens, D. and J. Meheus (2000), A Tableau Method for Inconsistency-adaptive Logics. In Automated
Reasoning with Analytic Tableaux and Related Methods, R. Dykhoff (ed.),. Berlin: Springer Verlag,
2000, pp. 127-172.
Batens, D. and J. Meheus (2001), Short Cuts and Dynamic Marking in the Tableau Method for Adaptive
Logic. Studia Logica 69, 221-248.
De Clercq, K. (200+), Two New Strategies for Inconsistency-adaptive Logics. Logic and Logical
Philosophy, in print.
Dowe, P. (1992), Wesley Salmon's Process Theory of Causality and the Conserved Quantity Theory.
Philosophy of Science 59, 195-216.
Dowe, P. (1995), Causality and Conserved Quantities: A Reply to Salmon. Philosophy of Science 62,
321-333.
Hempel, C. (1965), Aspects of Scientific Explanation and Other Essays in the Philosophy of Science. New
York: Free Press.
Hempel, C. (1966), Philosophy of Natural Science. Englewood Cliffs: Prentice-Hall.
Meheus, J. (2000), An Extremely Rich Paraconsistent Logic and the Adaptive Logic based on It. In
Frontiers of Paraconsistent Logic, D. Batens, C. Mortensen, G. Priest and J. P. Van Bendegem
(eds.), Baldock: Research Studies Press, pp. 189-201.
Salmon, W. (1984), Scientific Explanation and the Causal Structure of the World. Princeton, New Jersey:
Princeton University Press.
Salmon, W. (1993), The Value of Scientific Understanding. Philosophica 51, 9-19.
Salmon, W. (1994), Causality without Counterfactuals. Philosophy of Science 61, 297-312.
Salmon, W. (1998), Causality and Explanation. New York: Oxford University Press.
Kitcher, P. (1981), Explanatory Unification. Philosophy of Science 48, 507-531.
Kitcher, P. (1989), Explanatory Unification and the Causal Structure of the World. In Scientific
Explanation, P. Kitcher and W. Salmon (eds.), Minneapolis: University of Minnesota Press, 1989,
pp.283-306.
Van Fraassen, B. (1980), The Scientific Image. Oxford: Clarendon Press.
Weber, E. (1998), The Practical Functions and Epistemology of Causal Beliefs. Communication and
Cognition 31, 297-32.
JOHN D. NORTON
Abstract: Newtonian cosmology, in its original form, is logically inconsistent. I show the inconsistency
in a rigorous but simple and qualitative demonstration. "Logic driven" and "content driven" methods of
controlling logical anarchy are distinguished.
* This paper is a revised and updated version of John D. Norton, "A Paradox in Newtonian Cosmology",
PSA 1992, volume 2, pp. 412-420. The designation "II" of the title is to distinguish the versions since
some revisions are substantial. It is reprinted with the kind permission of the Philosophy of Science
Association.
185
J. Meheus (ed.), Inconsistency in Science, 185-195.
© 2002 Kluwer Academic Publishers.
186 JOHN D. NORTON
task of mastering the scientific technicalities let alone disentangling its logical
structure.'
My purpose here is to present an instance of a logically inconsistent theory
which is:
a real and significant piece of science"debated most recently in the primary
scientific joumalliterature of the 1950s;
indisputably logically inconsistent in the traditional strict sense that both
propositions A and not-A can be derived within the theory; and
technically so transparent that the inconsistency can be displayed essentially
without equations.
This instance is presented with an invitation to apply your own favorite analysis of
logical inconsistency in scientific theories in order to see how well your analysis fits.
The basic result to be developed here is that one can combine standard theorems in
Newtonian gravitation theory to conclude that
(1) The net gravitational force on a test mass at any point in space is F,
where F is a force of any nominated magnitude and direction.
Thus the theory is logically inconsistent, since we can prove within the theory that
the force on a test mass is both some nominated F and also not F, but some other
force.
In order to derive (1) from the postulates of Newtonian gravitation theory, we need
essentially only those properties of the Newtonian gravitational field which can be
represented in a simple lines of force picture. The essential properties which we
shall need are shown in Figure 1 and are:
, My own case study of the inconsistency of the old quantum theory of black body radiation (Norton
1987) is a good example, unfortunately. Compare with Smith 1988 and Brown 1990.
A PARADOX IN NEWTONIAN GRA VIT ATION THEORY II 187
The intensity of the gravitational force on a test mass is given by the density of
the lines of force and the direction of the force by the direction of these lines.
The lines of force can never cross.
The lines of force may only terminate in a source mass.
The total number of lines of force terminating in a source mass is proportional to
the mass of the source.
Lines offorce
cannot cross and Gravitational force
may only terminatc on test mass fixed
by direction
and density
of lines of force.
Notice that these properties already capture a very significant part of Newtonian
gravitational theory. For example, they are sufficient to establish that the
gravitational force exerted by a source mass on a test mass must diminish with the
inverse square of the distance between them in a three dimensional space. 2
To derive the inconsistency (I) within Newtonian cosmology, we first need two
theorems readily demonstrable within the lines of force picture.
Theorem 1
A spherical shell of source masses exerts no net force on a test mass located at
any position within the shell.
To see this, imagine otherwise, that is, that there is a net field within the sphere. (See
Figure 2.) The lines of force of this field must respect spherical symmetry. This
2 To see this, consider the spherically symmetric field of the mass. The same total number of lines of
force penetrate any sphere centered on the mass. But the area of the sphere increases with the square
of its radius. Therefore the intensity of the lines of force on the sphere's surface diminishes with the
inverse square of the radius. Since this intensity gives us the magnitude of the gravitational force on a
test mass located on the surface of the sphere, this force diminishes with the inverse square of distance
from the source mass.
188 JOHN D. NORTON
uniquely determines lines of force that lie radially in the shell and cross at the center.
Since there is no source mass at the center, this crossing is not allowed. Therefore
there can be no field within the shell and no net gravitational force on a test body
within it.
Theorem 2
A spherically symmetric source mass distribution has the same external field as a
point source of the same mass.
To see this, note that a field is fully specified if we fix its total number of lines of
force and require it to be spherically symmetric about some point. (See Figure 3.)
Only allowed
since lines of configuration
force cross of lines of
at center force
In this case, both external fields have the same number of lines of force, since
their sources have the same mass. Again both fields must be arranged spherically
symmetrically about the center of their source masses. Therefore both external fields
are the same.
Since Newtonian gravitation theory is a linear theory, we can compute the net
gravitational force on a test mass as the sum of the forces exerted by all the
individual source masses. To find the net gravitational force on a test mass in
Newtonian cosmology, we may consider the infinite source mass distribution
divided into finite parts. Each part exerts some (possibly vanishing) force on the test
mass and the net force is simply the sum of these forces. It turns out that dividing
upthe sources masses in different ways, in this case, can lead to a different final net
A PARADOX IN NEWTONIAN GRAVITATION THEORY II 189
force . In particular, for any nominated force F on a test mass, it turns out that we can
always find a way of dividing the source masses of Newtonian cosmology so that
the resultant net force is F.
Both
fields
arc
spherica ll y
sym metric .
How this may be done is summarized in Figure 4. First we consider a test mass
within a Newtonian cosmology and nominate a force F of arbitrary magnitude and
direction. We then divide the source mass distribution into a sphere surrounded by
concentric shells. The sphere is chosen so that the test mass sits on its surface. The
sphere's position and size are determined by the requirement that the net force
exerted by the sphere on the test mass be F. Theorem 2 ensures that we can always
find a sphere of suitable size and position to satisfy this requirement. 3 The test mass
lies within the concentric shells of the remaining source masses. Therefore, from
Theorem I, each of these shells exerts no net gravitational force on the test mass.
Summing, the total force exerted by all source masses-the sphere and the
concentric shells-equals the arbitrarily chosen F and we have recovered (I) stated
above.
3 The point is intuitively evident, but I give the details for zealots. From the theorem, the force due to the
sphere is the same as the force due to a corresponding point source of equal mass located at the
sphere's center. Thus, by placing the center of the sphere in some arbitrarily nominated direction from
the test mass, we can fix the direction of F arbitrarily. Similarly we can set the magnitude of F
arbitrarily by choosing an appropriate radius for the sphere. It turns out that the force exerted by the
sphere grows in direct proportion to its radius, so that all magnitudes are available. To see this linear
dependence, note that the force exerted by the corresponding point source of Theorem 2 grows in
direct proportion to its mass and decreases as the inverse square of the radius of the sphere. However
the mass of the sphere grows in direct proportion to its volume, that is, as the radius cubed.
Combining the two dependencies, we recover the direct proportion of the force and the radius.
190 JOHN D. NORTON
Step 1
Nominate force F • • test
of arbitrary size mass
and direction F
on test mass.
Step 2
Add in a sphere of
source masses
that exerts a net
gravitational force F
on the test mass.
Step 3
Add in all
remain ing
source
masses in
concentric
pherical
shells.
Each shell
exerts no net
gravitational
force on the
test mass.
Figure 4. Proof that net gravitational force on a test mass is any arbitrarily
nominated force within Newtonian cosmology
a foil to assist his development of relativistic cosmology in the mid 191Os. 5 Research
into Newtonian cosmology did not cease with the advent of relativistic cosmology.
In the early 1930s, Milne and McCrea discovered that, in certain aspects, Newtonian
cosmology gave a dynamics identical to that of the relativistic cosmologies. This
result engendered a tradition of research in neo-Newtonian cosmology in which the
inconsistency had eventually to be addressed.
Within these many analyses of Newtonian cosmology, there seem to be three
types of responses by physical theorists to the inconsistency of Newtonian
cosmology:
They are unaware of the inconsistency and derive their results without
impediment. This was Newton's response when Bentley outlined the problem to
him in their celebrated correspondence; Newton simply denied there was a
problem!6 Milne and McCrea's early papers make no reference to the problem
even though they use the very construction of Figure 4 to arrive at a non-uniform
force distribution. (See Milne 1934 and McCrea and Milne 1934.)
They are aware of the inconsistency but ignore the possibility of deriving results
that contradict those that seem appropriate. This seems to be the case with
Narlikar (1977, 109-110), and certainly with Arrhenius (1909, 226) whose
diagnosis is that the paradox "only proves that one cannot carry out the
calculation by this method".7
They find the inconsistency intolerable and seek to modify one or other of the
assumptions of the theory in order to restore its consistency. (See Seeliger 1895,
1896; Einstein 1917, §1; Layzer 1954.) As my survey (Norton 1999) shows, of
those who explicitly address the problem, this is by far the most common
response. At one time or another, virtually every supposition of Newtonian
cosmology has been a candidate for modification in the efforts to eliminate the
inconsistency. These candidates include Newton's law of gravitation, the
uniformity of the matter distribution, the geometry of space and the kinematics
of Newton's space and time itself.
In all three cases, logical anarchy is avoided. In the first two cases, however, it is
not at all clear how it is avoided. At first glance, it would seem that the physical
theorists avoid logical anarchy by the simple expedient of ignoring it! Philosophical
work in logical inconsistency presupposes that something more subtle may really be
guiding the avoidance of logical anarchy and that it may be controlled by quite
principled methods. Most of these analyses implement what I shall call "logic driven
control of anarchy". Logical anarchy is avoided by denying or restricting use of
certain standard inference schemes within a non-standard or paraconsistent logic.
A difficulty of this approach is that it is hard to recover it from the actual
practice of physical theorists who do work with logically inconsistent theories.
Typically, it is hard to discern any principled approach to the control of logical
anarchy by such theorists. One certainly does not find explicit recourse to a
modification of something as fundamental and universal as basic schemes of logical
5 For a survey of work on the problem up to the end of the I 920s, see Norton 1999.
6 See Norton 1999 (Section 7.1) for details.
7 For further discussion see Norton 1999 (Section 7.2).
192 JOHN D. NORTON
8 See Smith 1988 for an account of a content driven approach to the control of anarchy.
9 Arrhenius (1909; quoted in Norton 1999. Section 7.1) mounts exactly this argument.
10 The indeterminateness of gravitational force on any given test body still remains. By appropriate
selection of fO' the force of a test body can still be set at any designated force. This choice however
now fixes the gravitational force distribution throughout space and thus the forces on all other bodies.
Malament (1995) has described this condition most clearly. This condition raises another issue of
interest elsewhere. It is usually assumed that the reformulation of Newton's theory in terms of a
potential field does not alter the physical content of the theory, or at least not its observable content.
We see in this cosmological case that it does. The reformulation does not admit many force
distributions and thus many observable motions that the original version did admit.
II This remarkable agreement resides in the following. Let R be the distance separating two different,
arbitrarily chosen masses in the Newtonian cosmology. Then the time dependence of R agrees exactly
with the time dependence of the scale factor (usually written as R also) in the Robertson-Walker
cosmologies of general relativity.
A PARADOX IN NEWTONIAN ORA VITA TION THEORY II 193
The difficulty with the content driven control of anarchy sketched above is that it
appears to be a completely ad hoc maneuver. What justifies ignoring all but one
preferred member of a set of conclusiqns derived validly from postulates? One
program would be to seek this justification in the logic driven control of anarchy.
Perhaps if we impose the restrictions of one or other non-standard logic upon
Newtonian cosmology, then we will recover the apparently ad hoc rules of the
content driven approach. This is an interesting possibility worth pursuing, but it is
not the only one.
We can also justify the strategy of content driven control without tinkering with
something as fundamental and universal as logic. If we have an empirically
successful theory that turns out to be logically inconsistent, then it is not an
unreasonable assumption that the theory is a close approximation of a logically
consistent theory which would enjoy similar empirical success. The best way to deal
with the inconsistency would be to recover this corrected, consistent theory and
dispense with the inconsistent theory. However, in cases in which the corrected
theory cannot be identified, there is another option. If we cannot recover the entire
corrected theory, then we can at least recover some of its conclusions or good
approximations to them, by means of meta-level arguments applied to the
inconsistent theory.
The clearest example is the case of homogeneous, isotropic cosmologies with
static mass distribution. In any such cosmology-Newtonian or otherwise,
symmetry considerations will require the vanishing of the net gravitational force on
a test mass. Thus, when we use these symmetry considerations to exclude all but
vanishing forces on a test mass in a static, Newtonian cosmology, we are in effect
saymg:
We know that this cosmology is inconsistent. However, we expect that a small
modification would eliminate the inconsistency and in the resulting, corrected theory
the only force derivable would be the one satisfying the symmetry requirement, that is,
the vanishing force.
In many cases, we might even guess what this corrected theory might be. Seeliger
(1895, 1896) noticed that merely adding an exponential attenuation factor to the
inverse square law of gravity would suffice. At very large distances only, the force
of gravity would fall off faster with distance than the inverse square. Because of the
enormous empirical success of Newton's theory, such attenuation factors must have
nearly negligible effects within our solar system, so that unambiguous empirical
determination of the factor is extremely difficult, as Seeliger found.
The case of neo-Newtonian cosmologies is similar and harbors a surprise. The
selection of the canonical fields can be justified by displaying a corrected theory in
which these canonical fields arise without inconsistency. Because of the agreement
over dynamics, one might imagine that the corrected theory would simply be general
relativity. But, in response to an earlier version of this paper, David Malament
(1995) showed me and the readers of Philosophy of Science that the correction can
be effected in the simplest and most desirable way imaginable. That is, we should
like the correction merely to consist in the elimination of a superfluous assumption
194 JOHN D. NORTON
that plays no essential role in the theory in the sense that its elimination does not
alter the observable consequences of the theory. The eliminable assumption proves
to be the assumption that there are preferred inertial states of motion in space. It
amounts to the adoption of a kind of relativity of acceleration. The corrected theory
no longer portrays the motion of cosmc masses as deflections by gravitational
forces from the preferred inertial motions. Instead the free fall motions of the cosmic
masses are taken as primitive. Using techniques introduced by Cartan and Friedrichs
in the 1920s, one constructs a gravitation theory that is observationally identical
with the original theory. Its novelty is that the free fall motions are represented by a
curvature of the affine structure of spacetime in a way that is strongly analogous to
the corresponding result in general relativity. The removal of the inconsistency of
Newtonian cosmology proves to be a natural and compelling path to the notion that
gravitation is to be associated with a curvature of the geometrical structures of
spacetime. 12
In sum, my proposal is that the content driven control of anarchy can be justified
as meta-level arguments designed to arrive at results of an unknown, consistent
correction to the inconsistent theory. The preferred conclusions that are picked out
are not interesting as inferences within an inconsistent theory, since everything can
be inferred there. Rather they interest us solely in so far as they match or
approximate results of the corrected, consistent theory.13
REFERENCES
Norton, J. D. (1999), The Cosmological Woes of Newtonian Gravitation Theory. In The expanding
Worlds of General Relativity, H.Goenner, J. Renn, J. Ritter and T. Sauer (eds.), Einstein Studies, vol.
7, pp. 271-323.
Seeliger, H. (1895), Ober das Newton'sche Gravitationsgesetz. Astronomische Nachrichten 137, 129-136.
Seeliger, H. (1896), Ober das Newton'sche Gravitationsgesetz. Bayerische Academie der Wissenschajien.
Mathematische-Naturwissenschajiliche Klasse 126, 373-400.
Smith, J. (1988), Inconsistency and Scientific Reasoning. Studies in History and Philosophy of Science
19,429-445.
NANCY J. NERSESSIAN
1. INTRODUCTION
I This, of course, is a question that can be asked only from the perspective of historical hindsight.
Maxwell thought he had derived the laws of a Newtonian system. That is, he believed he had
incorporated electrodynamics into the Newtonian framework by providing a macro-level analysis of
phenomena that would later be explained at the micro-level in terms of Newtonian forces in the
aether.
197
1. Meheus (ed.), Inconsistency in Science, 197-211.
© 2002 Kluwer Academic Publishers.
198 NANCY J. NERSESSIAN
after he had already derived the equations by some other means. 2 The standard
account paints Maxwell as searching for a way to make the set of equations for
closed circuits formally consistent for open circuits as well:
However, I have been arguing for some time that the best sense that can be made of
the historical records is to take the modeling seriously as generative of Maxwell's
representation and that leaves philosophers with the task of explaining how it is a
productive form of reasoning.
In constructing the mathematical representation of electromagnetic field concept,
Maxwell created several fluid vortex models, drawing from the source domains of
continuum mechanics and of machine mechanics. On my analysis, these analogical
domains served as sources for constraints used in interaction with those provided by
the target problem to create imaginary analogs that served as the basis of his
reasoning. As with analogical modeling generally, in conceptual innovation such
modeling often requires recognition of potential similarities across, and integration
of information from, disparate domains. An abstraction process I call "generic
modeling" is key in such reasoning. In viewing a model generically, the reasoner
takes the specific instantiation as representing features common to a class of
phenomena. For example, in reasoning about a triangle, one often draws or imagines
a concrete representation. However, to reason about what it has in common with all
triangles, in the reasoning process one needs to understand the instantiation as
2 See, e.g., Chalmers 1973, Chalmers 1986, Duhem 1902, and Heimann 1970. For discussion of a more
salient role for the models see, e.g., Berkson 1974, Nersessian 1984a, Nersessian 1984b, Nersessian
1992, Nersessian 200+, and Siegel 1991.
INCONSISTENCY, GENERIC MODELING, AND CONCEPTUAL CHANGE 199
lacking specificity in the size of angles and the length of sides. It was through
generic abstraction that, e.g., Newton could reason about the commonalities between
the motions of planets and of projectiles. In the paper I will focus on how generic
modeling enabled Maxwell to tolerate physical and formal inconsistencies in his
derivation of the field equations through model-based reasoning. The analysis
establishes that, contrary to the standard account, considerations of formal
consistency play little role in Maxwell's analysis.
We will not go through the details of these maneuvers in Part I except to point
out two noteworthy derivations. First, he derived an expression relating current
density to the circulation of the magnetic field around the current-carrying wire
j = Ij47t curl H (equation 9, p. 462). This equation agreed with the differential form
of Ampere's law he had derived in the earlier paper (1855-6, p. 194). The derivation
given here still did not provide a mechanism connecting current and magnetism.
Second, he established that in the limiting case of no currents in the medium and a
unified magnetic permeability, the inverse square law for magnetism could be
derived. Thus the system agreed in the limit with the action-at-a-distance law for
magnetic force (equations 19-26, p. 464-66).
In Part I Maxwell was able to provide a mathematical representation for
magnetic induction, paramagnetism, and diamagnetism. Although the system of a
medium filled with infinitesimal vortices does not correspond to any known physical
system, Maxwell used mathematical properties of a single vortex to derive formulas
for quantitative relations consistent with the constraints on magnetic systems
discussed above. The four components of the mechanical stress tensor, as interpreted
for the electromagnetic medium, are: F = [H(lj47t div IlH)] + [lj87tIl(grad H2)]_
[IlH X Ij47tcurl H) - [gradpd (equations 12-14, p. 463). By component they are
(i) the force acting on magnetic poles, (ii) the action of magnetic induction, (iii) the
force of magnetic action on currents, and (iv) the effect of simple pressure. The last
component is required by the model-it is the pressure along the axis of a vortex-
but had not yet been given an electromagnetic interpretation. Note that we call the
contemporary version of this equation the "electromagnetic stress tensor" even
though it is now devoid of the physical meaning of stresses and strains in a medium.
A mechanical inconsistency in the hydrodynamic model led Maxwell to a means
of representing the causal relationships between magnetism and electricity. He
began Part II by stating that his purpose was to inquire into the connection between
the magnetic vortices and current. Thus he could no longer simply consider just one
generic vortex in his analysis. He admitted a serious problem with the model: he
"found great difficulty in conceiving of the existence of vortices in a medium, side
by side, revolving in the same direction" (p. 468). Figure 2 is my drawing of a cross
INCONSISTENCY, GENERIC MODELING, AND CONCEPTUAL CHANGE 201
section of the vortex model as described by Maxwell. To begin with, at the places of
contact among vortices there will be friction and, thus, jamming. Further, since they
are all going in the same direction, at points of contact they would be going in
opposite directions. So, in the case where they are revolving at the same rate, the
whole mechanism should stop. Maxwell noted that in machine mechanics this kind
of problem is solved by the introduction of "idle wheels". OR this basis he proposed
to enhance his imaginary model by supposing that "a layer of particles, acting as idle
wheels is interposed between each vortex and the next" (p. 468). He stipulated that
the particles would revolve in place without slipping or touching in direction
opposite to the vortices. This is consistent with the constraint that the lines of force
around a magnetic source can exist for an indefinite period of time, so there can be
no loss of energy in the model. He also stipulated that there should be no slipping
between the interior and exterior layers of the vortices, making the angular velocity
constant. This constraint simplified calculations, but is inconsistent with the
mechanical constraint that the vortices have elasticity, and would be eliminated in
Part III.
The model is now a hybrid constructed from two source domains: fluid dynamics
and machine mechanics. In ordinary mechanisms, idle wheels rotate in place. In the
model this allows representation of action in a dielectric, or insulating, medium. To
represent current, though, the idle wheels need to be capable of translational motion
in a conducting medium. Maxwell noted that there are mechanical systems such as
the "Siemens governor for steam-engines" have idle wheels that can translate out of
place. Throughout Part II, he provided analogies with machinery as specific
mechanical interpretations of the relations he had derived between the idle wheel
particles and the fluid vortices to establish that there are real physical systems that
instantiate the generic relations. The constraints governing the relationships between
electric currents and magnetism are modeled by the relationships between the
vortices and the idle wheels, conceived as small spherical particles surrounding the
vortices. Figure 3 is Maxwell's own rendering of the model. The diagram shows a
202 NANCY J. NERSESSIAN
cross section of the medium. The vortex cross sections are represented by hexagons
rather than circles, presumably to provide a better representation of how the particles
are packed around the vortices. The major constraints that need to be satisfied are
that (1) a steady current produces magnetic lines of force around it, (2) commence-
ment or cessation of a current produces a current, of opposite orientation, in a nearby
conducting wire, and (3) motion of a conductor across the magnetic lines of force
induces a current in it.
The analysis began by deriving the equations for the translational motion of the
particles in the imaginary system. There is a tangential pressure between the
surfaces of spherical particles and the surfaces of the vortices, treated as
approximating rigid pseudospheres. Mechanically, conceived as such these vortices
are inconsistent with the geometrical constraints of the vortices in Part I which
require the vortices to be elastic, but this would not be addressed until the analysis of
static electricity in Part III. Maxwell noted that the equation he derived for the
average flux density of the particles as a function of the circumferential velocity of
the vortices p = 1/2p curl v (equation 33, p. 471) is of the same form as the equation
relating current density and magnetic field intensity j = 1/4rr curl H (equation 9).
This is the form of Ampere's law for closed circuits he had derived in Part I. All
that was required to make the equations identical was to set 'p' , the quantity of
particles on a unit of surface, equal to 1/2. He concluded that "it appears therefore,
according to our hypothesis, an electric current is represented by the transference of
the moveable particles interposed between the neighboring vortices" (p. 471). That
is, the flux density of the particles represents the electric current density.
INCONSISTENCY, GENERIC MODELING, AND CONCEPTUAL CHANGE 203
We can see how the imaginary system provides a mechanical interpretation for
the first constraint. Current is represented by translational motion of the particles. In
a conductor the particles are free to move but in a dielectric (which the aetherial
medium is assumed to be) the particles can only rotate in place. In a nonhomoge-
neous conducting medium, different vortkes would have different velocities and the
particles would experience translational motion. They would experience resistance
and waste energy by generating heat, as is consistent with current. A continuous
flow of particles would thus be needed to maintain the configuration of the magnetic
lines of force about a current. The causal relationship between a steady current and
magnetic lines of force is captured in the following way. When an electro-motive
force, such as from a battery, acts on the particles in a conductor it pushes them and
starts them rolling. The tangential pressure between them and the vortices sets the
neighboring vortices in motion in opposite directions on opposite sides-thus
capturing polarity of magnetism-and this motion is transmitted throughout the
medium. The mathematical expression (equation 33) connects current with the
rotating torque the vortices exert on the particles. Maxwell went on to show that this
equation is consistent with the equations he had derived in Part I for the distribution
and configuration of the magnetic lines of force around a steady current (equations
15-16, p. 464).
Maxwell derived the laws of electromagnetic induction in two parts because each
case is different mechanically. The first constraint in the case of electromagnetic
induction ((2) above) can be reformulated as: a changing (non-homogeneous)
magnetic field induces a current in a conductor. The analysis began with finding the
electro-motive force on a stationary conductor produced by a changing magnetic
field. This first case corresponds, for example, to induction by switching current off
and on in a conducting loop and having a current produced in a nearby conducting
loop. A changing magnetic field would induce a current in the model as follows. A
decrease or increase in the current will cause a corresponding change in velocity in
the adjacent vortices. This row of vortices will have a different velocity from the
next adjacent row. The difference will cause the particles surrounding those vortices
to speed up or slow down, which motion will in tum be communicated to the next
row of vortices and so on until the second conducting wire is reached. The particles
in that wire will be set in translational motion by the differential electro-motive
force between the vortices, thus inducing a current oriented in direction opposite to
the initial current, which agrees with the experimental results. The neighboring
vortices will then be set in motion in the same direction and the resistance in the
medium will ultimately cause the translational motion to stop, i.e., the particles will
only rotate in place and there will be no induced current. Maxwell's diagram (Figure
3) illustrates this mechanism. The accompanying text (p. 477) tells the reader how to
animate the drawing for the case of electromagnetic induction by the action of an
electro-motive force of the kind he had been considering. The mechanism for
communicating rotational velocity in the medium accounts for the induction of
currents by the starting and stopping of a primary current.
In deriving the mathematical relationships, Maxwell used considerations about
the energy of the vortices. The mathematics is too complex to include in detail here.
The general procedure he followed was to derive the equations for the imaginary
204 NANCY J. NERSESSIAN
connexion existing in nature" (p. 486). Rather it is "a mode of connexion which is
mechanically conceivable and easily investigated, and it serves to bring out the
actual mechanical connexions between the known electro-magnetic phenomena"
(ibid.). His analysis thus far still had not shown that action in a medium is essential
for transmitting electromagnetic forces. Eor this he would have to show that there is
a necessary time delay in the transmission of the action and be able to calculate the
velocity of the transmission in the medium. He concluded in the summary of his
results that he had at least shown the "mathematical coherence" of the hypothesis
that electromagnetic phenomena are due to rotational motion in aetherial medium.
The paper had not yet provided a unified theory of electric and magnetic actions
because static electricity was not yet incorporated into the analysis. Thus, all
Maxwell claimed was that the "imaginary system" gives "mathematical coherence,
and a comparison, so far satisfactory, between its necessary results and known facts"
(p.488).
Provides Provides
Generic Generic
Identification Identification
Rotational
Elastic Stress
Motion
Generic
bstraction
Elastic tension
Rotation of Magnetic
between vortices and
Vortices
idle wheel particles
There are few existing drafts of Maxwell's published papers, but fortunately
there are a couple of pages of a draft pertaining to this derivation. 3 These reveal
clearly that Maxwell had some confusion about how to think of current and charge
without the medium. In the draft equation for current, Maxwell wrote the
components of "electric resistance", i.e. t4e electro-motive force required to keep the
current flowing through a conductor, as pointing in the opposite direction, as would
have been the case in the mechanical medium, but in the published paper these are
written correctly. In the draft equation for "statical electricity" the components are
written with the negative sign as above, but Maxwell's handwriting leads to the
interpretation that he struggled with this.4 For the first two components, the equals
sign is three times the length customary in Maxwell's handwriting and actually
blends into the minus sign. Only for the final component is the equals sign of regular
length and clearly separated from the minus sign. As noted above, it also written
with the minus sign in the published paper. 5
Maxwell never discussed this inconsistency and then in the Treatise, again
without discussion, the inconsistency is gone. Although it is just speculation on my
part, given how Maxwell collects all of his equations in Part III of the published
paper, it is hard to imagine that he did not notice the inconsistency. I interpret his
keeping the Coulomb equation p = - div D (equation G, p. 561) in 1864 because he
had not figured out how to conceive of charge generically, that is to abstract it from
the specific mechanism through which he had been able initially to represent it. That
the initial representation of static electricity was difficult for him to work out in the
1861-2 analysis is indicated by the fact that he submitted the part of the analysis
pertaining to it eight months after the work on magnetism and electromagnetic
induction was published. In the Treatise charge is abstracted from the notion of
stress and reactive force due to accumulating particles and treated generically as
elastic stress from the flow of electricity through a medium, with orientation in the
direction of flow. In conduction the current flow is unimpeded and in electrostatics,
stress is created at points of discontinuity, such as where a charging capacitor and a
dielectric meet. The generic notion of charge as associated with elastic stress is
compatible with Faraday's field notion of charge, but was to cause difficulties in
comprehending Maxwell's for those who took the customary action-at-a distance
notion that charge is associated with a particle. As H. A. Lorentz noted in a
discussion of the need for a clear separation of field and charge, "Poincare mentions
a physicist who declared that he understood the whole of Maxwell's theory, but that
he still had not grasped what an electrified sphere was!" (Lorentz 1891, 95, my
translation from the Dutch).
We have seen here that both mechanical inconsistencies in the versions of the
models and a formal inconsistency in the system of equations were tolerated in
scientific reasoning in a significant qeative episode for the sake of making
representational progress in a domain. Although Maxwell's mechanical model is not
an over-all consistent mechanical system this did not impede his analysis in deriving
the electromagnetic field equations. The model was understood to represent the
dynamical relations between the idle wheels and vortices in generic form. That is,
although a concrete mechanism was provided, in the reasoning process, the idle
wheel-vortex system is taken to represent the class of dynamical systems having
certain abstract relations in common. This class includes electric and magnetic
interactions on the assumptions of Maxwell's treatment. Thus, e.g. in the analysis of
electromagnetic induction, the idle wheel-vortex mechanism is not the cause of
electromagnetic induction; it represents the causal structure of that kind of process.
From the specific hybrid model, Maxwell formulated the laws of the electromag-
netic field by abstracting the dynamical properties and relations continuum-
mechanical systems, certain machine mechanisms, and electromagnetic systems
have in common. In their mathematical treatment these common dynamical
properties and relations were separated from the specific instantiation of the model
through which they had been rendered concrete. Thus the underlying inconsistencies
in the model could be ignored. The generic mechanical relationships within the
imaginary system served as the basis from which he abstracted a mathematical
structure of sufficient generality that it could represent causal processes in the
electromagnetic medium without requiring knowledge of specific causal
mechanisms. However we have also seen that at an intermediate stage of
development specific features of the model seemed to figure so strongly in his
thinking that he introduced a formal inconsistency in the set of equations that was
only eliminated several years later-again through generic modeling.
Returning to the broader question of conceptual change raised at the outset, we
can see how the abstraction process of generic modeling enabled the construction of
a field concept of electromagnetic forces from mechanical systems. A central
problem for understanding conceptual (or representational) change in science is how
is it possible to create new conceptual systems that both derive from and are
inconsistent with existing systems. I see this as a dimension of the "problem of
incommensurability" and I have argued extensively in previous work for the
"continuous but not simply cumulative" nature of the development of representa-
tional systems in a domain. Generic modeling enabled Maxwell to construct a
system of abstract laws that when applied to the class of electromagnetic systems
yield the laws of a dynamical system that is non-mechanical, i.e., one that is
inconsistent with the laws of the mechanical domains from which its mathematical
structure was abstracted. It was only in the development of the special theory of
relativity that mechanics and electromagnetism together are given a consistent
representational structure. Our analysis needs to end with Maxwell, except to note
that abstraction through generic modeling played a significant role in H. A. Lorentz'
INCONSISTENCY, GENERIC MODELING, AND CONCEPTUAL CHANGE 211
College of Computing and School ofPublic Policy, Georgia Institute of Technology, USA
REFERENCES
Berkson, W. (1974), Fields of Force: The Development ofa World View from Faraday to Einstein. New
York: John Wiley & Sons.
Chalmers, A. F. (1973), Maxwell's Methodology and his Application of it to Electromagnetism. Studies
in the History and Philosophy of Science 4,107-164.
Chalmers, A. F. (1986), The Heuristic Role of Maxwell's Mechanical Model of Electromagnetic Phenomena.
Studies in the History and Philosophy of Science 17,415-427.
Duhem, P. (1902), Les theories electriques de J. Clerk Maxwell: Etude historique et critique. Paris: A.
Hermann & Cie.
Duhem, P. (1914), The Aim and Structure of Physical Theory. New York: Atheneum.
Harman (Heimann) (1995), (ed.), The Scientific Letters and Papers of James Clerk Maxwell. Vol. II,
Cambridge: Cambridge University Press.
Heimann, P. M. (1970), Maxwell and the Modes of Consistent Representation. Archive for the History of
Exact Sciences 6,171-213.
Maxwell, J. C. (1861-2), On Physical Lines of Force. In Scientific Papers. Vol. I, W. D. Niven (ed.),
Cambridge: Cambridge University, pp. 451-513.
Maxwell, J. C. (1864), A Dynamical Theory of the Electromagnetic Field. In Scientific Papers. Vol. I, W.
D. Niven (ed.), Cambridge: Cambridge University, pp. 526-597.
Nersessian, N. J. (1984a), Aether/or: The Creation of Scientific Concepts. Studies in the History and
Philosophy of Science 15,175-212.
Nersessian, N. J. (l984b), Faraday to Einstein: Constructing Meaning in Scientific Theories. Dordrecht:
Martinus NijhofflKluwer.
Nersessian, N. J. (1992), How do Scientists Think? Capturing the Dynamics of Conceptual Change in
Science. In Cognitive Models of Science, R. Giere (ed.), Minneapolis: University of Minnesota,
pp.3-44.
Nersessian, N. J. (200+), Abstraction via Generic Modeling in Concept Formation in Science. Technical
Report 94/22, Cognitive Science Series, Georgia Institute of technology, to appear in Correcting the
Model: Abstraction and Idealization in Science, M. R. Jones and N. Cartwright (eds.), Amsterdam:
Rodopi.
Siegel, D. (1991), Innovation in Maxwell's Electromagnetic Theory. Cambridge: Cambridge University
Press.
INDEX
Batens, D., 43, 119, 129, 131, Carnot's theorem, 151, 154, 156, 160
136-137,142,144-145,147-149, Cartan, 194
153, 159-160, 163, 169-170, Cartwright, N., 8-9, 11-13, 16-19,23,
173-174,183-184 31,90,102,211
Bayes' theorem, 85 Cat, 1., 31
Beets, F., 149 category theory, 55
Bell inequalities, 86, 102 celestial mechanics, 5
Bell, J. L., 54-57, 86 Chalmers, A. F., 198,211
Belnap, N. D., 138, 143, 148 Chamberlain, T. C., 27, 31
Benferhat, S., 144, 149 Cherniak, c., 22, 31
Bentley, 191 Chiappin,1. R. N., 69, 77
Berkson, W., 198,211 Chuaqui, R., 65-67, 72, 77-78, 85,
Bernoulli, 53 103,112,117-118
Beth, 9 chunking, 123, 126, 151-152
B6ziau, 1.- y., 68, 77 circumscription, 147
biogeography, 89-90 Clark, P., 154, 163
black-body radiation, 97, 105, 108, classical logic, 55, 63, 94-95,
110-111 105-106,114, 129-130, 136, 138,
Bloor, D., 57 151, 166, 182, 197
Bogen, J., 17,24,31 classical mechanics, 11-12, 16, 24,
Bohr, N., 24, 38,41,90,97, 101, 36,89, 107, 109
105-110, 113, 115-117, 122, Clausius, R., 134-136, 151, 154,
124-125 156-161,163
bottom, 138 CLuN, 137, 146-147, 160, 174-177,
Bourbaki, N., 71-72, 77 183
Boyd, R., 85 CLuNs, 146-147
Boyer, C. B., 46, 57 Cn systems, 138
Brandom, R., 21-22, 33,144,150 Cohen, R. S., 32
Brewer, R., 33 coherence, 15-16,22, 108, 131-132,
Brouwer, 136 134,205
Brown, B., 31, 81, 86-88, 99,102, Collins, B., 33
105, 107-112, 115-117, 124, 128, Compton, 38
151,163,185, 194 concept formation, 197
Brownian motion, 21, 151 conceptual change, 197,210
Brush, S., 21, 31 conceptual dynamics, 130, 147
Bueno, 0., 59-60, 65-66, 68, 70, confirmation, 7, 87,107,120
76-77,85,102,112-113,117 consilience, 22, 88
calculus consistency
differential, 52 internal, 16, 20
infinitesimal, 44-45, 56, 60, 68, mutual, 7,12,16-17,26,123
105,122,129 of nature (the world), 3, 8, 18, 131
logical, 8-10, 12, 16 proof theoretic sense, 8
Campbell, D. T., 28, 31 semantic sense, 9
Cantor, 55, 59, 72-75, 129 constructive empiricism, 97
Carnot, S., 39,134-136,151, constructive empiricists, 82, 87-88,
154-158, 160-161 97
INDEX 215
1. F. Hallyn (ed.): Metaphor and Analogy in the Sciences. 2000 ISBN 0-7923-6560-7
2. 1. Meheus (ed.): Inconsistency in Science. 2002 ISBN 1-4020-0630-6