What is Thought?

  • Eric B. Baum
Bradford (MIT Press): 2004. 495 pp. $40, £25.95 | ISBN: 0-262-02548-5
Thinking ahead: is thought a program that decodes sensory information so we can make decisions? Credit: D. TEEL/CORBIS

“What is thought?” is not a new question. For Aristotle, thought was what the soul does, and for Descartes it was the unequivocal evidence of one's existence. For Eric Baum, a US expert in machine learning, it is a computer program. This is not a superficial assertion: Baum pursues the idea with elegance, clarity and considerable persuasion. His choice of book title was inspired by Edwin Schrödinger's book What is Life?, which was published before the discovery of DNA. Schrödinger felt that the answer to “What is life?” must lie within physics and chemistry, despite their apparent failure to deliver this at the time. Similarly, Baum argues that an answer to “What is thought?” must lie within computer science, despite its failure to produce convincing results so far under the guise of artificial intelligence.

It is important not to treat the idea that thought is a program in too superficial a fashion. Popular texts often include explanations such as “brain is like hardware and mind is like software”. Baum intends a level of sophistication far above this. The program to which he refers is one that extracts meaning from complex data. Thought for him is the process that ‘understands’ the complexities of the world. So a thought program is one that detects sensory information as a compact code.

For example, when walking down the road we may see a dog coming towards us. Is it friendly or not? We will need to take different actions depending on the answer. What thoughts are taking place in our heads and where do they come from? They involve a heavy dose of recalled experience and even just-learned information. We might once have been bitten by a pit-bull terrier, or just know from pictures or descriptions by others that if the dog is unmuzzled and untethered then it might be best to seek refuge behind a door, for example.

All the information we need to make our decision is there in the sensory world, but its meaning needs to be decoded from myriad other data that are not relevant to the key issue, danger in this case. There may be people, cars, other dogs, leaves carried on the wind and so on. But our thought decodes ‘danger’ and plans a reaction to it. Calling this thinking activity a ‘decoding and planning program’ requires several caveats. First, it is unlikely that a human could have written a program that could lead to a lifetime of thought. On the other hand, evoking an Almighty programmer is a recourse to mysticism that would leave us befuddled rather than enlightened. Baum's central point is that it is quite possible for programs to evolve, adapt and learn, making them more powerful than anything that a programmer can concoct.

A second caveat is that Baum does not suggest that someone is going to create an evolutionary, adaptive, learning program, put it into a robot and create a thinking object indistinguishable from a human being. Rather, Baum's argument is that a good (if not the best) way of understanding human thought is to analyse it as if it were a program. Artificial intelligence, in the past, was the product of programmers writing smart programs that do clever things, such as beating Garry Kasparov at chess. Baum goes beyond this by presenting clear explanations of what it is to extract the compactly coded information in the world using the simplest possible program; what it is for such a program to come into being through a process of evolution and adaptation; and what it is for a program to learn both over several generations and during daily life. Baum calls this ‘Occam's razor’ programming, stressing that the simplest program model is likely to have the best powers of explanation.

The third caveat is my own. One must understand that Baum's explanation is a metaphorical or functional one — it is not a material model of the brain. Even though some of his computational models are ‘neural networks’ (a technology loosely inspired by the operation of brain cells), Baum does not suggest that a careful analysis of the brain will reveal that it is structured like a general-purpose computer on which the meaning-extraction programs run. His point is that meaning computation is a powerful explanatory metaphor for what the brain, its complex neurons, its chemical transmitters, its muscular engagement with the world and its highly adapted and impenetrable architecture do in an abstract sense. Although Baum does attempt to answer questions about what kind of program is needed for awareness, consciousness and will, he does so at a metaphorical level and does not address what the material brain does that makes it appear to a computer theoretician to be running these programs.

An important part of the book is devoted to the considerable flak that artificial intelligence scientists attracted in the heyday of this topic. Baum gives a reasoned response to John Searle's claim that no program can ‘understand’ the world, and to Roger Penrose's contention that conscious insight lies outside the logic that can be achieved by computation. The essence of the response is that these objections were made before ideas such as evolution, adaptation and compact decoding had become part of the idea of computation. Baum joins in the criticism of what is now called GOFAI, or ‘good old-fashioned artificial intelligence’, so this is a splendid book for discovering what is new. It will enthral some computer scientists and provoke some philosophers. And it should engage general readers who wish to enjoy a clear, understandable description of many advanced principles of computer science.