Documente Academic
Documente Profesional
Documente Cultură
J. R. LUCAS
II7
P H I LO SO P H Y
trivial,question.But it is enough,enoughto showthatthemachine
is notthesameas a mind.True,the machinecan do manythings
thata humanmindcannotdo: but ifthereis ofnecessitysomething
that the machinecannotdo, thoughthe mindcan, then,however
trivialthe matteris, we cannotequate the two, and cannothope
everto have a mechanicalmodelthatwilladequatelyrepresent the
mind.Nor does it signify that it is onlyan individualmachinewe
have triumphed over:forthetriumph is notoveronlyan individual
machine,but over any individualthat anybodycares to specify-
in Latin quivis or quilibet. not quidam-and a mechanical model
of a mindmustbe an individualmachine.Althoughit is truethat
any particular"triumph"of a mind over a machinecould be
"trumped"by anothermachineable to producethe answerthe
firstmachinecould not produce,so that "thereis no questionof
triumphing simultaneously overall machines",yetthisis irrelevant.
What is at issue is not the unequalcontestbetweenone mindand
all machines,but whethertherecouldbe any,single,machinethat
coulddo all a mindcan do. For themechanistthesisto holdwater,
it mustbe possible,in principle,
to producea model,a singlemodel,
whichcan do everything the mindcan do. It is like a game.I The
mechanisthas firstturn.He producesa-any, but onlya definite one
-mechanical model of the mind. I point to somethingthat it
cannotdo, but the mindcan. The mechanistis freeto modifyhis
example,but each timehe does so, I am entitledto lookfordefects
in the revisedmodel.If the mechanistcan devise a modelthat I
cannotfindfaultwith,his thesisis established:if he cannot,then
it is not proven:and since-as it turnsout-he necessarily cannot,
it is refuted.To succeed,he mustbe able to producesomedefinite
mechanicalmodelof the mind-any one he likes,but one he can
specify,and willstickto. But sincehe cannot,in principlecannot,
produceany mechanicalmodelthat is adequate, even thoughthe
pointof failureis a minorone,he is boundto fail,and mechanism
mustbe false.
Deeper objectionscan stillbe made. Gddel'stheoremappliesto
deductivesystems,and humanbeingsare not confinedto making
only deductiveinferences. Gddel's theoremapplies only to con-
sistentsystems,and one may have doubts about how far it is
permissibleto assume that humanbeingsare consistent.Gddel's
theorem appliesonlyto formalsystems, and thereis no a prioribound
to humaningenuity whichrulesout thepossibility ofourcontriving
somereplicaofhumanitywhichwas not representable by a formal
system.
I For a similartype of argument,see J. R. Lucas: "The Lesbian Rule";
PHILOSOPHY, July I955, PP. 202-6; and "On not worshippingFacts"; The
Philosophical Quarterly, April i958, p. I44.
ii8
MINDS, MACHINES AND GODEL
Human beingsare not confinedto makingdeductiveinferences,
and it has been urgedby C. G. Hempel,and HartleyRogers;that
a fairmodelof the mindwouldhave to allow forthe possibility of
makingnon-deductive inferences,and thesemightprovidea way
of escapingthe Gddel result.Hartley Rogersmakes the specific
suggestionthat the machineshould be programmedto entertain
variouspropositions whichhad not been provedor disproved,and
on occasionto add themto its listofaxioms.Fermat'slast theorem
or Goldbach's conjecturemightthus be added. If subsequently
theirinclusionwas foundto lead to a contradiction, theywouldbe
droppedagain, and indeedin those circumstances their negations
wouldbe addedto thelistoftheorems. In thissortofwaya machine
mightwellbe constructed whichwas able to produceas truecertain
formulaewhichcould not be provedfromits axiomsaccordingto
its rulesof inference.And therefore the methodof demonstrating
themind'ssuperiority overthemachinemightno longerwork.
The construction ofsucha machine,however,presentsdifficulties.
It cannot accept all unprovableformulae,and add them to its
axioms,or it willfinditselfacceptingboththeGbdelianformulaand
its negation,and so be inconsistent.Nor wouldit do if it accepted
the firstof each pair of undecidableformulae,and, havingadded
thatto itsaxioms,wouldnolongerregarditsnegationas undecidable,
and so wouldneveracceptit too: forit mighthappenon thewrong
memberof the pair: it mightaccept the negationof the G6delian
formularatherthan the Godelianformulaitself.And the system
constitutedby a normalset of axioms with the negationof the
Gb5delian formulaadjoined,althoughnotinconsistent, is an unsound
system,not admitting ofthenaturalinterpretation. It is something
like non-Desarguiangeometriesin two dimensions:not actually
inconsistent,but ratherwrong,sufficientlymuchso to disqualifyit
fromseriousconsideration. A machinewhichwas liableto infelicities
ofthatkindwouldbe no modelforthehumanmind.
It becomesclearthatrathercarefulcriteriaofselectionofunprov-
able formulae willbe needed.HartleyRogerssuggestssomepossible
ones. But once we have rulesgenerating new axioms,even if the
axiomsgeneratedare onlyprovisionally accepted,and are liable to
be droppedagainiftheyare foundto lead to inconsistency, thenwe
can set about doinga Gbdelon this system,as on any other.We
are in the same case as whenwe had a rulegenerating the infinite
set ofGodelianformulaeas axioms. In short,howevera machineis
designed,it mustproceedeitherat randomor accordingto definite
rules.In so faras its procedureis random,we cannotoutsmartit:
I In private conversation.
X I957, Vol. I,
Theoryof RecursiveFunctionsand EffectiveComputability,
pp. 152 ff.
II9
PHILOSOPHY
but itsperformance is notgoingto be a convincing parodyofintelli-
gent behaviour:in so far as its procedureis in accordancewith
definiterules,the Godel methodcan be used to producea formula
which the machine,accordingto those rules, cannot assert as
true,althoughwe, standingoutsidethe system,can see it to be
true.,
Godel's theoremapplies onlyto consistentsystems.All thatwe
can proveformallyis that if the systemis complete,then the
G6delianformulais unprovable-in-the-system. To be able to say
categorically thattheGddelianformulais unprovable-in-the-system,
and therefore true,we must not onlybe dealingwitha consistent
system, but be able to saythatit is consistent.
And,as Gbdelshowed
in his secondtheorem-a corollaryof his first-itis impossibleto
provein a consistent systemthatthatsystemis consistent. Thusin
orderto faultthe machineby producinga formula ofwhichwe can
say boththatit is trueand that the machinecannotproduceit as
true,we haveto be able to saythatthemachine(or,rather,its corre-
sponding formalsystem)is consistent; and thereis no absoluteproofof
this.All we can do is to examinethe machineand see if it appears
consistent.There always remainsthe possibilityof some incon-
sistencynot yet detected.At best we can say that the machineis
consistent,providedwe are. But by what rightcan we do this?
Godel's second theoremseems to show that a man cannotassert
his ownconsistency, and so HartleyRogerszarguesthatwe cannot
really use Godel's firsttheoremto counterthe mechanistthesis
unlesswe can say that"thereare distinctive attributes whichenable
a humanbeingto transcendthislast limitationand asserthis own
consistency whilestillremaining consistent".
A man's untutoredreactionif his consistency is questionedis to
affirm it vehemently: but this,in view of Godel's secondtheorem,
is takenby somephilosophers as evidenceofhisactualinconsistency.
ProfessorPutnam3has suggestedthathumanbeingsare machines,
but inconsistent machines.If a machinewerewiredto correspond to
an inconsistent system,thentherewouldbe no well-formed formula
whichit could not produceas true; and so in no way could it be
provedto be inferiorto a humanbeing. Nor could we make its
inconsistencya reproachto it-are not men inconsistenttoo?
Certainly womenare,and politicians;and evenmalenon-politicians
I G6del's originalproofapplies if the rule is such as to generatea primitive
MertonCollege,Oxford.
127