Speaking Minds Interviews with Twenty Eminent Cognitive Scientists EDITED BY PETER BAUMGARTNER AND SABINE PAYR

Speaking Minds

Interviews with Twenty Eminent Cognitive Scientists

EDITED BY PETER BAUMGARTNER

AND SABINE PAYR

INTRODUCTION

Why a Book of Interviews?

The idea for a collection of interviews with notable cognitive scientists originated as early as 1989, while we were in residence at the University of California at Berkeley, working as research scholars at the Institute of Cognitive Studies. Peter Baumgartner was preparing a book on educa-tional philosophy dealing specifically with the background of knowl-edge (Baumgartner 1993), and Sabine Payr was working on her doctoraldissertation on knowledge-based machine translation (Payr 1992). So we both had already been dealing with the problems of Artificial Intelligence and cognitive science, and we shared a focus on the philosophical and social questions raised by these fields.

Take It Apart and See How It Runs

When I was a graduate student at Pittsburgh in 1966, we worked through Quine's book Word and Object.1 That book was one of the few things in philosophy that made any sense to me at that time, apart, of course, from Hume. Several other graduate students whom I came to know quite well were utterly contemptuous of the later Wittgenstein in particular and of so-called ordinary language philosophy in general. Their scoffing was typically directed toward specific claims, such as the private language argument, and they generally had very good arguments to support their criticism. And apart from seeming willfully obscure,Wittgenstein was at best insensitive to the possible role of scientific data in philosophical questions and, at worst, openly hostile to science. Thus the skepticism about conceptual analysis as a method for finding out how things are in the world, including that part of the world that is the mind, was germinating and taking root while I was at Pittsburgh. In Oxford from 1966 to 1969, by contrast, Wittgenstein was revered; and having taken a Quinean position on the a priori and on intentionality, I ended up making a nuisance of myself in Oxford by swimming in the other direction. Paul [Churchland] was most helpful around this time,because he was very straightforward. He thought we ought to approach mind questions in basically the same scientific spirit with which one approached other questions of fact. So instead of flailing around aimlessly analyzing concepts, I had some sense of direction. And so I tried to figure out what I could do in the philosophy of mind if it was not conceptual analysis. Because I am a materialist, it finally occurred to me that there might be a lot in neuroscience that might help answer the questions I was interested in—for example, questions about reasoning, decision making,and consciousness.

Neural Networks and Commonsense

I would say that I have been in cognitive science my entire career, because I think a philosopher of science, someone who works on epistemology, is a cognitive scientist. For a long time, philosophers were the only cognitive scientists. Many philosophers pursued their interests in a nonempirical way, and I think that was unfortunate. But I have always been inclined to do my philosophy in a way that responds to empirical data, so I think I have been doing cognitive science for twenty years. But it is more rewarding now than it was twenty years ago, because there are so many interesting things coming from neuroscience and from Artificial Intelligence.

Cognition and Cultural Belief

I first got interested in language and thought when I took my master's degree in 1955 at UCLA, where I was introduced to the work of Edward Sapir1 by Harry Hoijer, an anthropological linguist. In summer of 1955, I met Alfred Schutz2 and extended my familiarity with his work. These contacts stimulated me to go outside of sociology.

In the spring of 1954,1 began a series of small research projects with Harold Garfinkel3 that led to a serious interest in everyday reasoning and social interaction. These activities influenced my dissertation research. My dissertation got a lot of people in sociology upset because I talked about a number of issues that were about language, social interaction, and reasoning among the aged. Without using the word cognition, I was concerned with the phenomenal world of the aged and at the time was stimulated by phenomenological interests in philosophy. .. . I

corresponded briefly with Alfred Schutz at that time. I found phenom-enological ideas very useful, but I was bothered eventually by the fact that there was not much empirical research attached to that tradition.My preoccupation with the local ethnographically situated use of language and thought in natural settings was not always consistent with this tradition.

In Defense of Al

Before there was a field called cognitive science, I was involved in it when I was a graduate student at Oxford. At that time, I knew no science at all.I had had a purely humanistic education as an undergraduate. But I was very interested in the mind and in the philosophy of mind. I was completely frustrated by the work that was being done by philosophers, because they did not know anything about the brain, and they did not seem to be interested. So I decided that I had to start to learn about the brain to see what relevance it had. I became an autodidact neuroscientist, with the help of a few professors.

What I found out was that people who knew about brains did not have a lot to say about the mind, either. In those days, unlike today, it was very hard to come across much of anything in neuroscience that had the ambition of addressing any of the philosophical questions about mind.

Cognitivism Abandoned

How does a philosopher get involved in cognitive science? What is the connection between your work and cognitive science? Well, I got into it even before there was cognitive science, through Artificial Intelligence, because I was teaching at MIT about 1962 or 1963. Students from what was then called "Robot Project"—that is,from Minsky's job—came and said that they had already solved or were solving the problems that philosophy was worried about, like understanding and knowing and so on. It seemed to me at the time that that was unlikely, but if they had, it was something a philosopher had better know about.

The Folly of Simulation

I was trained as a philosopher at Princeton. My first academic job was at MIT, about 1960, when all the Chomsky1 stuff was starting. There was a lot of linguistics around there, which I picked up a little. Then I spent a year visiting the University of Illinois, where I was involved in a program of Charles Osgood's.2 I had a position in a research program he was running. I talked a lot with graduate students about psychology,and so I got involved in doing some experiments. For fairly fortuitous reasons, I picked up some information about philosophy, linguistics,and psychology. Since a lot of it seemed to bear on the same set of problems, I have worked back and forth between those fields ever since. It turned out that my own career had elements of the various fields, which converged to become cognitive science, except computer science, about which I know very little.

Farewell to GOFAI?

When I was in graduate school, Bert Dreyfus was one of my teachers. I first took a seminar on Artificial Intelligence offered jointly by him and Michael Scriven. They disagreed quite seriously, and I was impressed by Dreyfus's side and got involved in helping to formulate his argument. I did not actually write on Artificial Intelligence for my dissertation, but I became interested in the topic and continued to work on it when I came to Pittsburgh.

Then, I was member of a group at CASBS—Center for Advanced Study in the Behavioral Sciences—at Stanford in 1979 that was devoted to philosophy and Artificial Intelligence. In the course of that year, my work concentrated further.

Embodied Minds and Meanings

Probably I should start with my earliest work and explain how I got from there to cognitive science. I was an undergraduate at MIT, where I was a student of both mathematics and English literature. There I had an opportunity to begin to learn linguistics with Roman Jakobson, Morris Halle, and Noam Chomsky.11 was primarily interested in working with Jakobson on the relationship between language and literature. And that is an interest that has stayed with me throughout my career. My undergraduate thesis at MIT was a literary criticism thesis, but it contained the first story grammar. It was an attempt to apply Chomskian linguistics to the structure of discourse. I took Propp's Morphology of a Folk Tale1 and recast it in terms of transformational grammar. That led to the study of story grammar in general.

Toward a Pragmatic Connectionism

I started out with a fascination for my own thought processes. I found myself preoccupied with things that were getting in the way of progress. So I started reading on scientific psychology, and I slowly came to realize that I could not come up with an account of what was going on in my head. This was back in high school already.

When I came to college, I had an excellent course from a behavionst who convinced me that you could actually study what happened, what governed people's behavior. I became extremely involved in this work. I went to a lab and started doing behavionstic experiments, under the influence of the idea that all this stuff I was thinking about my thought processes was irrelevant, that I should study behavior and finally find the laws of behavior. After about a year I started asking questions about what was governing this behavior. I did experiments with animals and wanted to know why they chose this stimulus and not another—because they did not remember, because they saw, because it was like the one before?

The Serial Imperative

The field of computer science did not exist in the fifties. Computer science as a discipline does not show up until the early or midsixties. Before that, computers were viewed as engineering devices, put together by engineers who made calculators, where the programming was done by mathematicians who wanted to put in mathematical algorithms. Consequently, there was not really an intellectual discipline of programming.On the other hand, there clearly was the ferment of computer science in cybernetics. The postwar world was clearly in intellectual ferment. I would almost use the word chaos, but not in the sense that it was in trouble; it was just bubbling over. As the number of scientists was fairly small, lots of things that are regarded today almost as separate fields were thrown together.

Gestalt Psychology Redux

My training in cognitive science began as a graduate student in psychology at UC San Diego. With Donald Norman and David Rumelhart we began doing work in cognitive science before there was actually any field by that name, in that we, as psychologists, were constructing a large scale computer model of how people understand language and retrieve information from their memories. That was in the early seventies. During that time we came to know a number of important people in computer science and Artificial Intelligence, because Don organized a series of conferences. He brought people down to talk to us. Those people included Allen Collins and Ross Quillian,1 and Gordon Bower and John Anderson, who were doing the HAM model at that time.2 We went up to Stanford and visited Roger Schank's3 laboratory. Danny Bobrow4 came down from Xerox PARC (Palo Alto Research Center), and a number of other people came, too.

Against the New Associationism

I got involved in cognitive science in two ways, because I worked for many years in mathematics as well as in philosophy. I am a recursion theorist as well as a philosopher. So the theory of Turing machines is one that I have known, as it seems to me, for all my adult life. In the late fifties I suggested a philosophical position that I named "functionalism."It became very popular. I no longer believe in it, to the regret of many of my former students. But the idea is that the mind or the mind-brain is, basically, a computing machine, and instead of focusing attention on the question of whether the substance of our mind is or is not material, we should be focusing attention, rather, on the different question of whether the organization of our mind is or is not the same as that of a Turing machine.

From Searching to Seeing

I did my graduate work in the study of memory and discovered that there were too few constraints on what we do. Around 1970 I started trying to understand long-term memory: how is knowledge stored in memory, and what is stored? One thing we knew from psychology was that mainly meaning is important. People can remember things that are meaningful and will forget things that are not. As a graduate student I worked in the area of mathematical psychology. I thought that the tools of mathematical psychology were by and large too weak. So I became a model builder: I began to think that computer simulation was an important way to go. When I tried to figure out what kind of information might be stored in memory, and that it would have to do with meaning,I began to look around for people who knew anything about meaning,who knew what meaning is like. And there I saw several things.

Ontology Is the Question

How did I get involved in cognitive science? Well, it happened over adozen years ago. The Sloan Foundation decided to fund major research projects in the new area of cognitive science, and they asked various people in related disciplines if they would be willing to come to meetings and give lectures. I was invited, and I accepted. That way I met a lot of other people, and I was asked to participate in the formation of a cognitive science group in Berkeley. It is one of those things where outside funding actually makes a difference to your life. The fact that a major foundation was willing to fund a group and provide us with travel expenses, clerical support, and computers made a big difference to my participation. So, the short answer is that I got involved by way of the Sloan Foundation.

The Hardware Really Matters

Cognitive science came late in my career. It represents a synthesis of many of my other interests. The essential question for me has always been how the brain works. In the process of trying to understand how the brain works, I have come to appreciate the complexity not only of the structure of the brain but also of its cognitive abilities. My feeling now is that the study of the brain is going to require collaboration between at least two major areas of science—the cognitive sciences on the one hand and the neurosciences on the other hand—to make a synthesis possible.

Technology Is Not the Problem

My original area is organization and management. I approached organization and management from the standpoint of decision making. I had training in economics and understood the economist's model of decision making. But that seemed to me very far from what was going on in organizations. So I tried to develop alternative theories, theories that I generally label by "bounded rationality." This is the view that people are not as rational as economists think. That led me into trying to understand human problem solving, because it is part of what is involved in decision making. This research started back in 1939 and went on through the forties and fifties.

The Myth of the Last Metaphor

I was introduced to computers in the early fifties, at Wayne University(now Wayne State University) in Detroit. A mathematics professor there decided to build a computer for the university. So we did build one. I was a graduate student then. We had to do everything: soldering, all the logic design, and the assembler language. That is how I began with computers.

After a long while, I participated in the design of a computer system for the Bank of America. That was in the late fifties and early sixties. The actual site of the work was Palo Alto, California, because the headquarters of the Bank of America is in San Francisco, and the Stanford Research Institute had done some preliminary work that we had to know about. That was, of course, very near Stanford University. I already knew Ed Feigenbaum, who was then at Berkeley.

Why Play the Philosophy Game?

When I was a graduate student, I became interested in Artificial Intelligence and cognitive science. From my point of view, there is not really a difference between cognitive science and Artificial Intelligence. So for me, becoming interested in Artificial Intelligence was in fact becoming interested in cognitive science.Could you explain why there is no difference? What is Artificial Intelligence, and what is cognitive science?

It is not that there is no difference. It is that one discipline, in a sense,includes the other. Both Artificial Intelligence and cognitive science today—some of my colleagues may disagree with me—take a kind of information processing approach to cognition. This approach can be

Computers and Social Values

Things for me began in college, with an interest in computers and language. In the very beginning, those were the two things that interested me. I actually spent a year studying linguistics in London before I went to MIT and got involved in computer science. So I came into my study of computing with an interest in combining language and computers. Then I found out that the work on Artificial Intelligence at MIT seemed to be an obvious match. I did my dissertation work at MIT, then I was on the faculty there for three years before coming here (to Stanford).

broadly described by the question, What is the brain doing as a computational device? What is interesting about what the brain does is that it does some interesting stuff with information. And as soon as we start talking about doing things with information we are talking about information processing, and a very good way to talk about information processing is to use the kind of languages and methods that computer scientists have been developing for a long time.

The Albatross of Classical Logic

I have always been interested in the issue of machine thinking. Let me show you something that I wrote in a student magazine published at Columbia University, Thinking Machines—A New Field of Electrical Engineering. Here are some of the headlines: "Psychologists Report Memory Is Electrical," "Electronic Brain Able to Translate Foreign Languages is Being Built," "Electronic Brain Does Research," "Scientists Confirm on Electronic Brain." Behind these headlines are the questions:

How will electronic brains or thinking machines work and affect our living? What is the role played by electrical engineers in the design of these devices? These are some of the questions I tried to answer in this article.Now look at the date of the article: 1950. That was six years before the term Artificial Intelligence was coined. I want to underscore the headline "Electronic Brain Able to Translate Foreign Languages Is Being Built" published in 1950. Forty years later, we still do not have the kind of computer that can really translate foreign languages. It shows how easy it was and still is to overestimate the ability of machines to simulate human reasoning.