A.I: The Path towards Logical and Rational Agents: Thinking Machines
4/5
()
About this ebook
This book is an introduction to the fundamentals of Artificial Intelligence its aim is to introduce A.I. through its history, its contemporary role in society and the economy. We then lift the hood to see how A.I. works and how we can create our very own rational A.I. agents
The topics of interest are as follows:
- A.I. today in 2017
- A.I role in society
- A.I. role in employment
- A.I. role in the economy
- Algorithms and types of A.I
- Modern Approaches to A.I
- What A.I. can do
- What A.I. can't do – for now
- Types of A.I. agents
- Understanding Environments
- Solving Problems with A.I
- Planning with A.I
- Building Rational and Logical Agents
- Applying A.I. through Bayesian and Decision Networks
In early chapters of this book we will learn how AI has reemerged over the last couple of decades due to a rise in complimentary enabler technologies and some surprising successes that have brought A.I. to the mainstream media’s attention.
Then we lift the hood to start our investigations into how AI systems work. We will learn that AI is powered by algorithms, which do the underlying heavy lifting but we will also get an understanding regards how the algorithms achieve their goals and we will glimpse the principles on which they are designed to function. For example, we will learn the importance as to whether an algorithm acts humanly or rationally
Also we will introduce the concept and learn about AI agents what they are and how they relate to artificial intelligence as a technology. We will consider how we make agents intelligent and by what means. We will then consider and learn about the various types of AI agents and have a look at their individual architecture and learn more about each of their functions and purpose. Hence we will learn how to build reflex-based, model-based, goal-based, and utility-based agents by learning how they work. Finally we will introduce learning agents and a radical view of an alternative architecture based upon natures evolutional model.
We will address how to use AI algorithms to solve problems. Hence we will learn how to construct our problem, by ensuring it is well-defined. We will also learn how a search tree works and its infrastructure and how to decompose a search algorithm to its four functional components. We will also learn how to understand the algorithm search performance and results by completeness, optimality, and space and time complexity.
Then we will learn about knowledge-based agents, what they are, how they are built and how we imbue them with knowledge. Furthermore we will introduce and learn about Logical agents where we use representations based on logic statements to allow the agent to perform acts of reason.
In addition we look at AI's role in Planning and why we require AI as a planning agent. We will see how a planning agent's infrastructure differs from problem solving agents in state, goals and actions. Furthermore we will learn the theory behind Bayesian networks and Conditional Probability Tables and how to build them and calculate probability. We will also learn how to use inference algorithms in Bayesian network.
Finally we look at the principles behind rational agents and the 6 constraints known as the Axioms of Utility Theory that any preference model must consider when behaving rationally. Using the knowledge accrued over the previous chapters we will construct a decision network and learn how to use it to derive probabilities of events and calculate maximum expected utility (MEU).
Read more from Alasdair Gilchrist
Google Cloud Platform an Architect's Guide Rating: 5 out of 5 stars5/5An Executive Guide to Identity Access Management - 2nd Edition Rating: 4 out of 5 stars4/5Spreadsheets To Cubes (Advanced Data Analytics for Small Medium Business): Data Science Rating: 0 out of 5 stars0 ratingsGoogle Cloud Platform for Data Engineering: From Beginner to Data Engineer using Google Cloud Platform Rating: 5 out of 5 stars5/5REST API Design Control and Management Rating: 4 out of 5 stars4/5A Concise Guide to Object Orientated Programming Rating: 0 out of 5 stars0 ratingsConcise Guide to OTN optical transport networks Rating: 4 out of 5 stars4/5Concise Guide to DWDM Rating: 5 out of 5 stars5/5Six Sigma Yellow Belt Certification Study Guide Rating: 0 out of 5 stars0 ratingsA Practical Guide Wireshark Forensics Rating: 5 out of 5 stars5/5GDPR for DevOp(Sec) - The laws, Controls and solutions Rating: 5 out of 5 stars5/5A Concise Guide to Microservices for Executive (Now for DevOps too!) Rating: 1 out of 5 stars1/5Supply Chain 4.0: From Stocking Shelves to Running the World Fuelled by Industry 4.0 Rating: 3 out of 5 stars3/5The Layman's Guide GDPR Compliance for Small Medium Business Rating: 5 out of 5 stars5/5Concise and Simple Guide to IP Subnets Rating: 5 out of 5 stars5/5Digital Success: A Holistic Approach to Digital Transformation for Enterprises and Manufacturers Rating: 0 out of 5 stars0 ratingsThe Certified Ethical Hacker Exam - version 8 (The concise study guide) Rating: 3 out of 5 stars3/5Google Cloud Platform - Networking Rating: 0 out of 5 stars0 ratingsTackling Fraud Rating: 4 out of 5 stars4/5PSD2 - Open Banking for DevOps(Sec) Rating: 5 out of 5 stars5/5A Last Minute Hands-on Guide to GDPR Readiness Rating: 0 out of 5 stars0 ratingsAn Introduction to SDN Intent Based Networking Rating: 5 out of 5 stars5/5Concise Guide to CompTIA Security + Rating: 3 out of 5 stars3/5FinTech Rising: Navigating the maze of US & EU regulations Rating: 5 out of 5 stars5/5SRS - How to build a Pen Test and Hacking Platform Rating: 2 out of 5 stars2/5The Concise Guide to SSL/TLS for DevOps Rating: 5 out of 5 stars5/5ChatGPT Will Won't Save The World Rating: 0 out of 5 stars0 ratingsWhy Industry 4.0 Sucks! Rating: 0 out of 5 stars0 ratingsManagement Accounting for New Managers Rating: 1 out of 5 stars1/5The Concise Guide to the Internet of Things for Executives Rating: 4 out of 5 stars4/5
Related to A.I
Titles in the series (3)
Robotics: from Mechanical to Sentient Machines: Thinking Machines, #1 Rating: 0 out of 5 stars0 ratingsMachine Learning: Adaptive Behaviour Through Experience: Thinking Machines Rating: 4 out of 5 stars4/5A.I: The Path towards Logical and Rational Agents: Thinking Machines Rating: 4 out of 5 stars4/5
Related ebooks
Artificial Intelligence: Machine Learning, Deep Learning, and Automation Processes Rating: 5 out of 5 stars5/5Machine Learning: Adaptive Behaviour Through Experience: Thinking Machines Rating: 4 out of 5 stars4/5Artificial Intelligence: How Machine Learning, Robotics, and Automation Have Shaped Our Society Rating: 5 out of 5 stars5/5Artificial Intelligence: The Complete Beginner’s Guide to the Future of A.I. Rating: 4 out of 5 stars4/5Artificial Intelligence: Learning about Chatbots, Robotics, and Other Business Applications Rating: 5 out of 5 stars5/5Artificial Intelligence Implementing AI for your Business Rating: 4 out of 5 stars4/5Artificial Intelligence: Data Analytics and Innovation for Beginners Rating: 5 out of 5 stars5/5Artificial Intelligence: Robots, Applications, and Machine Learning in a Nutshell Rating: 5 out of 5 stars5/5Artificial Intelligence: Understanding Business Applications, Automation, and the Job Market Rating: 0 out of 5 stars0 ratingsArtificial Intelligence in Business and Technology: Accelerate Transformation, Foster Innovation, and Redefine the Future Rating: 0 out of 5 stars0 ratingsDeveloping Analytic Talent: Becoming a Data Scientist Rating: 3 out of 5 stars3/5Introduction to TinyML Rating: 5 out of 5 stars5/52062: The World that AI Made Rating: 5 out of 5 stars5/5Robotics: from Mechanical to Sentient Machines: Thinking Machines, #1 Rating: 0 out of 5 stars0 ratingsNeural Networks: A Practical Guide for Understanding and Programming Neural Networks and Useful Insights for Inspiring Reinvention Rating: 0 out of 5 stars0 ratingsArtificial Intelligence: Thinking Machines and Smart Robots with Science Activities for Kids Rating: 5 out of 5 stars5/5The AI Generation: Shaping Our Global Future with Thinking Machines Rating: 4 out of 5 stars4/5MINDFUL AI: Reflections on Artificial Intelligence Rating: 0 out of 5 stars0 ratingsThe Fourth Age: Smart Robots, Conscious Computers, and the Future of Humanity Rating: 3 out of 5 stars3/5Succeeding with AI: How to make AI work for your business Rating: 0 out of 5 stars0 ratingsArtificial Intelligence: A Guide for Thinking Humans Rating: 4 out of 5 stars4/5Artificial Intelligence: Understanding A.I. and the Implications of Machine Learning Rating: 3 out of 5 stars3/5Artificial Intelligence in Practice: How 50 Successful Companies Used AI and Machine Learning to Solve Problems Rating: 0 out of 5 stars0 ratingsTech Trends in Practice: The 25 Technologies that are Driving the 4th Industrial Revolution Rating: 0 out of 5 stars0 ratingsA.I. Hacked: A Practical Guide to the Future with Artificial Intelligence Rating: 0 out of 5 stars0 ratingsHow to Create Machine Superintelligence (Second Edition) Rating: 0 out of 5 stars0 ratings
Business For You
Becoming Bulletproof: Protect Yourself, Read People, Influence Situations, and Live Fearlessly Rating: 4 out of 5 stars4/5Capitalism and Freedom Rating: 4 out of 5 stars4/5Crucial Conversations Tools for Talking When Stakes Are High, Second Edition Rating: 4 out of 5 stars4/5Crucial Conversations: Tools for Talking When Stakes are High, Third Edition Rating: 4 out of 5 stars4/5Law of Connection: Lesson 10 from The 21 Irrefutable Laws of Leadership Rating: 4 out of 5 stars4/5Just Listen: Discover the Secret to Getting Through to Absolutely Anyone Rating: 4 out of 5 stars4/5The Richest Man in Babylon: The most inspiring book on wealth ever written Rating: 5 out of 5 stars5/5Nickel and Dimed: On (Not) Getting By in America Rating: 4 out of 5 stars4/5Emotional Intelligence: Exploring the Most Powerful Intelligence Ever Discovered Rating: 5 out of 5 stars5/5Robert's Rules Of Order Rating: 5 out of 5 stars5/5Collaborating with the Enemy: How to Work with People You Don’t Agree with or Like or Trust Rating: 4 out of 5 stars4/5Set for Life: An All-Out Approach to Early Financial Freedom Rating: 4 out of 5 stars4/5Lying Rating: 4 out of 5 stars4/5The Intelligent Investor, Rev. Ed: The Definitive Book on Value Investing Rating: 4 out of 5 stars4/5Summary of J.L. Collins's The Simple Path to Wealth Rating: 5 out of 5 stars5/5Leadership and Self-Deception: Getting out of the Box Rating: 4 out of 5 stars4/5Your Next Five Moves: Master the Art of Business Strategy Rating: 5 out of 5 stars5/5Confessions of an Economic Hit Man, 3rd Edition Rating: 5 out of 5 stars5/5Tools Of Titans: The Tactics, Routines, and Habits of Billionaires, Icons, and World-Class Performers Rating: 4 out of 5 stars4/5Buy, Rehab, Rent, Refinance, Repeat: The BRRRR Rental Property Investment Strategy Made Simple Rating: 5 out of 5 stars5/5High Conflict: Why We Get Trapped and How We Get Out Rating: 4 out of 5 stars4/5The Five Dysfunctions of a Team: A Leadership Fable, 20th Anniversary Edition Rating: 4 out of 5 stars4/5How To Pay Off Your Mortgage in 5 Years Rating: 5 out of 5 stars5/5The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers Rating: 4 out of 5 stars4/5Robert's Rules of Order: The Original Manual for Assembly Rules, Business Etiquette, and Conduct Rating: 4 out of 5 stars4/5Financial Words You Should Know: Over 1,000 Essential Investment, Accounting, Real Estate, and Tax Words Rating: 4 out of 5 stars4/5The Catalyst: How to Change Anyone's Mind Rating: 4 out of 5 stars4/5
Reviews for A.I
1 rating0 reviews
Book preview
A.I - alasdair gilchrist
An Introduction to Artificial Intelligence
Chapter 1 – Introducing Artificial Intelligence
A.I. in today’s world
A.I. hype or reality
Chapter 2 -A.I. Conundrum
A.I. - the jobs thief
A.I. and the future of jobs
A.I. – in the office
A.I. Big-Bang
A.I. and the economy
Chapter 3 - A.I. in real world
Algorithm and inferred knowledge
Algorithms – the building blocks
The master algorithms
Common Sense Reasoning
A.I. the Ethical and Moral Paradox
A.I. the legal dilemma
Chapter 4 - Technical fundamentals of A.I.
Approaches to A.I.
What can today's A.I. systems do?
Chapter 5 - AI Algorithm functions and principles
Human Action
Thinking Humanly
Thinking Rationally
Acting Rationally
So what are intelligent agents?
Types of Agents
Chapter 6 – AI Agents and their Environments
Agent architectures
Table based agent
Percept based agent or reflex agent
The model-based reflex agent
Goal-based agents
Utility based agents
Learning agents
Subsumption Architecture
Chapter 7 - Problem Solving with AI
Search Tree Algorithms
Search Algorithm Infrastructure
Measuring problem solving performance
Uninformed Search Strategies
Informed Search
A* Search Algorithm
RBFS (recursive best-first search)
Chapter 8 - Knowledge-based Reasoning
Knowledge-based agents
Building a Knowledge Base
Agents that act logically
Chapter 9 – Planning with AI
Effective Heuristics
Planning Graphs
Chapter 10 - Dealing with Uncertainty
Constructing Bayesian networks
Inference in Bayesian networks
Chapter 11 - Rational Agents
Decision networks
Learning Agents
Chapter 1 – Introducing Artificial Intelligence
In this opening chapter we will learn how AI has reemerged over the last couple of decades due to a rise in complimentary enabler technologies such as the internet and cloud technologies that have delivered data in abundance. We will learn how and why A.I. requires such abundance of raw data and how businesses are approaching the issue. We will also see some early but surprising headline successes that brought A.I. to the mainstream media’s attention. As a result, we will learn why there is so much hype surrounding AI and why it has become the darling of the press. We will also learn some of the basic realities and its history of cycles of overblown optimism.
A.I. in today’s world
Presently, in 2017 we are undoubtedly witnessing the crest of an Artificial Intelligence wave of hype but also of unsurpassed levels of progress asA.I. has become pervasive throughout our lives. The last decade has seen unprecedented levels of progress through innovation and the introduction of disruptive technologies that have changed the way we work, live and play. Most of the algorithms have A.I. at their core however many of these new technologies have come about due to the timely maturing of several enabler technologies.
The first wave of enabler technologies came with the birth and explosive growth of the internet and social media. By late 1990 and early 2000 the rate of uptake for both was staggering and for once data in the form of images and text content was in abundance. This sudden freely accessible and easily downloadable source of data was a huge boon to A.I. as it overcame the issues of procuring suitable quantities of test and training data to feed the algorithms.
In addition a second wave of new technologies quickly followed, such as pervasive and affordable wireless communications, the ubiquitoussmartphone and the highly complementary Internet of things (IoT), all of which enabled products,services and devices to provide utility to the public and generate even more data. Arguably the most important precursor technology, which has boosted A.I. development over the last decade and a half has been the emergence of Cloud computing.
Cloud services emerged in the early tomid-2000s, and the providers initially brought us Infrastructure as a Service (IaaS), Software as a Service (SaaS), and even Platforms as a Service (PaaS), which revolutionized how small, medium and enterprise companies managed their data centers, hosted their applications and integrated their IT departments into the business. For Cloud services made application development, application hosting and importantly data storage, through providing pay-as-you-use models for compute, memory and storage very affordable. This meant that even startups could realistically develop and launch cloud hosted applications without any upfront capital investment as they could simply pay for more immediately provisioned resources as and when required. These benefits of course were also available and affordable to research institutes and Universities who now had available to them the vast compute, storage, memory and network resources that previous generation of scientists could only have dreamed about.
Cloud computing played a massive role in the resurgence of A.I. over the last decade, because A.I. has at its core a need for data, lots of data, and that was previously a major constraint upon research. With Cloud services such as Amazon Web Services, Microsoft’s Azure and Google Cloud, data storage became extremely affordable with prices for storage in early 2017 at $0.021 per GB for standard storage and even cheaper for infrequent access or glacial storage.
During the latter half of the 2000s the emergence of this cheap storage led to the concept of Big Data, cloud storage repositories, or data lakes. However, storing large data sets is one thing, extracting meaningful value from them quite another. This is due to companies not having access to the required skills and analytical engines that are necessary, whether that be in batch or more importantly, in real-time streaming, to process the raw data. Analysing vast data sets is not a trivial task and you need to know what you are looking for before you can extract and transform the data. The insights do not just manifest out of the ether, they need to be discovered within the data. Furthermore not all data is the same even within the same business unit of a company. For example in manufacturing some data's value has a long life so can be extracted from historical records other data though has a very short life span and needs to be captured in a timely manner, to gain the operational process knowledge. For those reasons only around 1% of data being captured and hoarded by companies in 2016 was ever analysed.
Cloud service providers supply the storage mechanisms to create vast repositories of data that companies previously treated as a waste product. This made storing, supporting and maintaining extremely large data sets of historical and real-time data feasible even for small to medium businesses. Fuelled by a belief that data would somehow transform their business through predictive analytics and the benefits of the IIoT (Industrial Internet of Things), companies started to hoard data. The problem was however that most of the data went unanalysed. Despite this most companies still considered it more expensive and risky to dispose of data than manage data retention strategies and processes, their philosophy is store everything, throw nothing away, it might be worth something to somebody someday.
Perhaps unsurprisingly, as we moved into the early years of the 2010s this data retention policy paid off, with the emergence of an associated technology to A.I. – it is arguable whether it is still a subset of A.I. – of machine learning. Thankfully that vast repository of storage need not go to waste as data is the fuel that A.I. depends on. But if A.I. required vast quantities of data, then it was nothing compared to the voracious appetite of ML with it deep neural networksand supervised training.
Around 2012, the much studied and often maligned artificial neural networks (ANN) and deep neural networks in particular started to produce some staggering results. Deep neural networks are extensions (DNN) of artificial neural networks (ANN) just with considerably more hidden layers of neurons, hence the term relates to the network's depth. DNN's are closely associated with strong A.I. and the quest for Artificial General Intelligence (AGI), which is the search for human like reasoning, learning and knowledge reusability across a wide scope of disciplines i.e. the ability to apply what has been learned during a previous task and apply what is relevant to a new different task. This is difficult because up until February 2017 ANN were considered to be only usable for a specific task as the neuron network had to learn through iterations of supervised learning. Prior to DeepMind's breakthrough – they were able to retrain a neural network so it retained its original knowledge and circuits, while simultaneously learning new skills and tasks - an ANN's required to be completely retrained using the new learning data sets and test criteria. This was because attempting to repurpose an ANN already trained on a task would cause the neural network to start losing its established connections in a process called catastrophic forgetting, which resulted in the ability to perform the original task being lost.
A.I.'s machine learning algorithms require vast data sets in order to gain insightful correlations and predictions, which might not have been intuitive during the data preparation process. Indeed sometimes the cleansing and preparation of the data revealed correlations but more often was the main obstacle to analysis. Deep neural networks to a large degree have loosened that constraint as many work on an unsupervised or semi-supervised method. This lowers the barriers to entry for many companies that wish to use A.I. because they have the data, lots of data, sitting at rest in storage doing nothing but they don't have the resources or skills to process it. For these companies, A.I. algorithms could well be the answer to the problem of extracting value from the depths of their data lakes.
As we have seen A.I. provides searches that can assist in problem solving, planning and analytics, both with historical and real-time data. The larger the training data sets and learning time, the greater the potential rewards. However this requires not just storing large data sets but to have the processing and memory capabilities to realistically process the data to provide insight and reveal hidden correlations or anomalies within the data and that is now a realistic possibility for businesses of all sizes. Data is often referred to as the new oil such is its perceived value. This in many ways is true however we must not think that A.I. is going to be the answer to all business problems. It will not be a case of simply downloading an open-source A.I. engine and feeding it data – it is vastly more complicated than that.
What is true is that with A.I. the raw data is the oil in the machine, and that is why service providers and A.I. developers require vast repositories of data. IBM’s Watson when it played the TV quiz game ‘Jeopardy!’ was configured to have no internet access, but the IBM team provided Watson with millions of documents, including dictionaries, encyclopaedias – the entire Wikipedia, and other reference material that it could use to build its local knowledge store. Unsurprisingly witha knowledge-base containing 200 million pages of structured and unstructured content, consuming 4 terabytes of disk storage, that with such vast local knowledge Watson was the master at any general knowledge quiz.
The hardware for Watson includes a cluster of 2,880 POWER7 processor cores and 16 terabytes of RAM, with massively parallel processing capability.The utility of having so much RAM was that it facilitated that the entire knowledge base could reside within local memory making access times significantly quicker. When Watson was repurposed to provide amongst other A.I. tasks, medical diagnosis it again consumed vast tomes and tracts of data, not just historical health records and text books but also by reading thousands of newly published scientific papers per day – no human doctor could possibly have the time or motivation to achieve this mundane task. The thing is though, neither does Watson, because Watson doesn't read as we read, it has absolutely no idea what it is 'reading' it has no sense of the meaning or context, it is just data, lots and lots of raw data.
It is little wonder that within a year or so Watson could match or better general practitioners in medical diagnosis based on symptom and cause. Interestingly, when Watson worked in the far larger all-encompassing field of general cancer research, which is a massive field covering hundreds of types of cancer, Watson could still perform as well or better than specialist human doctors. Furthermore, Watson was a match for researchers and consultants even in their highly specific and very limited fields in providing correct diagnosis. Some critics will say that it's not intelligence it's not diagnosing a potential medical condition it is simply a battery of clever search engines with some probability algorithms thrown into to the mix to filter the results down to the most likely answer that matches all or most of the search criteria.
However, IBM’s Watson A.I. is not just some quick database, running smart algorithms to refine speed, efficiency and accuracy, Watson has other tricks up its virtual sleeves, and Artificial Intelligence is one of them.
A.I. hype or reality
Artificial Intelligence – is getting hyped so much it has managed to usurp the Internet of Things (IoT) as the new darling of the media. Indeed A.I. has been discussed in the media so much that like IoT it is even crossing over into the mainstream media. The hype regards this latest’s technology would have the public confused though as it is quite hard to differentiate whether it is an all-conquering process that will transform the world for good or the folly of humankind's own making, an evil manifestation that will ultimately destroy us all. Fortunately, A.I. is likely to be neither if history is anything to go by.
The first thing we have to understand about Artificial Intelligence is thatA.I. is not something new it has been around for a very long time. Furthermore, not only have most pensioners been through many similar A.I.'s surges in optimism and scaremongering they have also witnessed the sudden collapse in confidence, they've heard it all before. A.I. itseems goes through cyclical periods of unbounded hype followed by long periods, decades even, of disillusionment, which is often referred