AI and Human Thought and Emotion

Sam Freed

• Publisher: Auerbach Publications
• Year: 2019
• Print ISBN: 9780367029296
• eBook ISBN: 9780429671197
• Edition: 1
• Topic: Computer Science
• Subtopic: Artificial Intelligence (AI) & Semantics
• Index: Computer Science

The field of artificial intelligence (AI) has grown dramatically in recent decades from niche expert systems to the current myriad of deep machine learning applications that include personal assistants, natural-language interfaces, and medical, financial, and traffic management systems. This boom in AI engineering masks the fact that all current AI systems are based on two fundamental ideas: mathematics (logic and statistics, from the 19th century), and a grossly simplified understanding of biology (mainly neurons, as understood in 1943). This book explores other fundamental ideas that have the potential to make AI more anthropomorphic.

Most books on AI are technical and do not consider the humanities. Most books in the humanities treat technology in a similar manner. AI and Human Thought and Emotion, however is about AI, how academics, researchers, scientists, and practitioners came to think about AI the way they do, and how they can think about it afresh with a humanities-based perspective. The book walks a middle line to share insights between the humanities and technology. It starts with philosophy and the history of ideas and goes all the way to usable algorithms.

Central to this work are the concepts of introspection, which is how consciousness is viewed, and consciousness, which is accessible to humans as they reflect on their own experience. The main argument of this book is that AI based on introspection and emotion can produce more human-like AI. To discover the connections among emotion, introspection, and AI, the book travels far from technology into the humanities and then returns with concrete examples of new algorithms. At times philosophical, historical, and technical, this exploration of human emotion and thinking poses questions and provides answers about the future of AI.

I
INTELLIGENCE IN COMPUTERS, HUMANS AND SOCIETIES

Chapter 1

Artificial Intelligence as It Stands

This chapter will present the current state of the artificial intelligence (AI) world: What types of AI exist, and who are the central figures who had the most influence on how the AI world understands itself.

1.1 About AI

1.1.1 AI’s Relation to Psychology, Cognitive Science, etc.

AI did not appear out of nowhere nor is it unrelated to other fields. Some would say that AI is “the intellectual heart of the cognitive sciences” (Wheeler, 2005), but strangely, AI predates cognitive science (at least under those names). The history is complex and need not concern us here in all its detail (Boden, 2008). This section will give the context needed for the rest of the book.

Psychology is a science, in the sense that it aims to understand a phenomena that is objectively out there is the world: Human thinking and behaviour. Psychology is usually dated either to Plato or to the mid-19th century. Human behaviour includes the pesky trait of insisting, using language, that there is such a thing as subjectivity and a personal point of view (Seth, 2010). Some, however would resist this idea that psychology is a science, in that it has precious few theories that produce useful predictions – in any case nothing like physics or biology. Thomas Kuhn, in his analysis of scientific paradigms, gave a possible solution to this question: Psychology may be today where chemistry was before Dmitri Mendeleev invented the “periodic table of the elements” and gave chemistry its first theoretical framework (Kuhn, 2012). Psychology does not have an overall theory (yet). That does not devalue the work of psychologists (who are grappling in the dark about the main questions of their discipline) – rather, it makes their work more heroic – they are collecting the myriad facts and observations that may allow for a theory of human subjectivity and behaviour to emerge some time in the future.

The interests of psychology are divided very roughly between scientific psychologists who want to explain the mind like physicists explain matter, and therapists who are interested in helping individuals, and tend to be more interested in individual cases than in scientific theories. These two are joined by many other sub-fields such as social psychology, who want to help people achieve goals, whether personal, social, economical, or political. These form elements of the social sciences.

The history of psychology is divided very roughly into three periods, with perhaps a fourth one starting about now (the 2010s).¹

“Classical” psychology (started in the mid-19th century) was done mainly in Germany and included people exploring their own subjectivity, along with electrical currents in nerves. It was an open and uncritical time, when every researcher who could afford to set up a laboratory and develop their own methodologies published books, and was either accepted or not by their peers. Some would argue that several of the schools of thought that flourished at that time were indistinguishable from one another other than by academic politics.

This era came to a close (especially in the United States) with the behaviouristic revolution, led by John B. Watson, from 1913 onwards. The idea of behaviourism is that any discussion of the subjective is suspect, since we can never agree on what is happening in the subjective realm, and also as scientists we want to explore the behaviour of all animals, not just humans, and we have no access to the subjectivity underlying the behaviour of birds, a research field of Watson’s. We will discuss Watson in a more detail below.

The cognitive revolution in psychology was a rebellion against the austerity of behaviourism. In getting rid of anything subjective or introspective behaviourism had pretty much banned any discussion of the mind. The cognitivists (as they were later named) rebelled against the idea that there can be no discussion of “mental processes” since the acquisition of language by infants cannot be explained by the simple dog-like processes of stimulus and response. This movement was spearheaded by Noam Chomsky (1959) and included people from a variety of disciplines. Interestingly, many of the same people were heavily involved in the development of computers. This is no coincidence: In a sense, cognitive psychology is based on the metaphor of “mind as machine” or more specifically of “mind as computer”.

A key player both in the world of computers and in the nascent cognitive psychology was Herbert A. Simon, who will be discussed in more detail in Section 1.6, since he probably was the most influential single person in the formative decades of AI (G. Solomonoff, 2016).

The fourth era in psychology which is arguably gaining steam is a full-fledged return to interest in the subjective. This movement goes under the banner of “consciousness science” (Seth, 2010).

Cognitive science is an unusual field, in that in many contexts it is not even seen as a separate discipline. Rather, it is a meeting place of several disciplines, where researchers exchange ideas and explore shared interests. The fields include psychology and neurology (of course), but also linguistics, cybernetics, electrical engineering, computer science, and our field, AI.

Strangely, in a sense cognitive science existed as an intellectual movement at least since the late 1950s, but it only got its name from the book titled “Cognitive Psychology”, published in 1967. AI was given its name in 1956, at a conference in Dartmouth. History is rarely tidy. We will return to both of these events.

1.1.2 What Are Intelligence, Consciousness, and Introspection

The question of the definition of intelligence is complex, and even the Stanford Encyclopedia of Philosophy, where one would expect to find definitions and discussions of difficult words, has no article on the topic. Even the idea of measuring intelligence without defining it is mind-boggling in its complexity (Hernández-Orallo, 2017). Instead of delving into defining intelligence, I will preliminarily use the following cobbled-together definition:

Intelligence is proficiency in the acquisition and application of knowledge.

Any discussion of intelligence is also a discussion of knowledge, and there are complexities here, but we can leave those to Section 6.7.

I will not attempt to define consciousness. I will simply follow widespread custom in using “consciousness” as the sum total of all our subjective experience, and hence, consciousness is accessible to us, by looking at our own experience. And looking at (and reporting on) our own subjective experience is a pretty good initial definition of introspection. So for now, introspection is how we look at consciousness, and consciousness is the totality of what is available to introspection. If the circularity of these preliminary definitions worries you, recall that even in mathematics, some terms (What is 1?) are left undefined. And our entire language, as defined in the dictionary, is utterly circular, since all words are forever defined using other words. This book is ultimately about technology, so we cannot afford to get bogged down in definitions. For better definitions of introspection, see Chapter 7. Human consciousness as such is not a topic in this book.

1.1.3 Defining and Viewing AI

AI is a strange field. It is young – the name was only made up in 1956. It is part of the computer revolution from the start, actually from before the official start of electronic computing during the Second World War (McCorduck, 2004).

In the summer of 1956, several young researchers, mostly still in graduate school, gathered in Dartmouth. It was no normal conference with a clear beginning and end; it was more of a walk-on walk-off informal gathering of like-minded people, all interested in computers. Many were interested also in the ongoing cognitive revolution in psychology and neighbouring fields. The most influential character in the gathering was Herbert A. Simon, already a professor at the time (see Section 1.6).

At that conference, a definition of AI was agreed. It may not be the most satisfying definition, but it stuck.

… the artificial intelligence problem is taken to be that of making a machine behave in ways that would be called intelligent if a human were so behaving.

(G. Solomonoff, 2016, p. 6)

An interesting result of this definition is that once you program something that qualifies as AI, like a chess program to beat the world champion, from the moment it works, and you have some time to catch your breath, it is no longer AI. Since it just worked, it has just been done by a computer, and therefore, it is no longer interestingly human, it’s just a “relatively sophisticated piece of programming” – and all the glory is gone. This is one of the reasons that university departments of AI are rare and getting rarer – AI is usually taught in departments of computer science, as a sort of “advanced programming”.

AI used to be taught (like many other fields) historically. But now it is most often taught using the book titled “Artificial Intelligence: A Modern Approach” (Russell & Norvig, 2013). The “modern” bit of the approach seems to be that AI is presented in that book purely as engineering, as a series of useful techniques, with no history at all.

Some psychologists and cognitive scientists see AI not as a technology, but as the theoretical wing of cognitive science (a bit like theoretical physics). By their lights, a theory in cognitive science should be a program, a piece of code, that produces a similar result to a human’s cognitive system. AI is (by these researchers) the “intellectual heart of cognitive science” (Wheeler, 2005).

As a matter of historical fact, AI approaches can broadly be separated into two-and-a-half groups, depending on the original inspiration driving them.

1.2 First Approach: Logic and Mathematics

The first approach to AI, based on logic (and mathematics), is that of explicitly specified knowledge. At its most primitive, the AI system knows certain facts, and some rules of inference. An early example of this approach was the “Logical Theorist” (Newell & Simon, 1956) which was aimed at proving theorems in logic. Later examples were the “expert systems” used widely in science and medicine (Shortliffe et al., 1984). In these systems, the rules of the domain on knowledge (say chemistry or medicine) are precoded in the system, and by adding some facts about a particular case, the system can (automatically, flawlessly, and quickly) infer everything that can be inferred using the rules based on the data.

This first approach of representing knowledge explicitly using rules and facts later evolved to handle uncertainty, whereby every fact is represented as a probability, and the rules therefore produce diagnoses (for example) with a specific certainty factor. Another extension of this approach allowed even rules to have a probability, like “if A and B then the chances of C are 90%”. The expert systems that used to deal with knowledge that was either clearly present or not were thereby generalised to handle uncertainties, without losing any of its mathematical rigour and explicit clarity. Another development allowed a more sophisticated use of statistics, known a Bayesian systems and later Bayesian networks (Boden, 2016).

Another (early and important) example of logical/mathematical AI is the algorithm used to play chess and other turn-taking board games. This was initially proposed by Alan Turing, in 1953 (Turing, 1953). The basic idea is that for every situation on the board, there are so many moves that are possible. That number may be (say for example) 5. So one could draw a tree of possible board situations, with the current situation on the top, and all the possible immediately subsequent board situations below. We can apply this idea of making a tree junction out of every board situation time and again, until we have a tree of all possible developments in the game up to a certain dept, see Illustration 1.1.

Illustration 1.1 A two-layer game tree.

Next, we use some procedure to evaluate all the “leaf” positions that will not be further expanded. This may be done crudely, as in adding some small integer for every piece “our” side has on the board, while deducting something similar for opponent pieces. A large number can represent...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Table of Contents
• Author
• 0 Introduction
• Part I Intelligence in Computers, Humans and Societies
• Part II An Alternative: Ai, Subjectivity, and Introspection
• Part III Getting Practical
• Bibliography
• Index