Tuesday, 6 July 1999

Roger Penrose: The Emperor's New Mind (1989)

Edition: Oxford University Press, 1989
Review number: 283

It is easy to lose track of what The Emperor's New Mind is about. It is intended as an examination of the claims of 'strong AI' from a fairly sceptical viewpoint (as the title indicates). The style aims at popular science, but it is difficult to follow and occasionally, I suspect, dauntingly full of complex mathematics (as a mathematician myself, I find this a little difficult to

The strong AI position is basically that it is possible for a sufficiently complex computer program to mimic the workings of an intelligent mind. In computer science terms, this is more precisely expressed by saying that the workings of the mind are algorithmic or computable in nature. Penrose's book naturally starts, then, with an explanation of these terms, together with a discussion of related ideas by Turing (the famous Turing test, and the Turing machine, which is one of the more accessible ways to define computability). He also talks about results showing that there are concepts which are not algorithmic in nature: the halting problem (there is no algorithm to determine in general if a computer program will ever complete its task) and Gödel's incompleteness theorem (which implies that there is no computer program which can mechanically prove every mathematical result about arithmetic). This first section ends with a digression into the lambda calculus, an alternative formulation of computability, which is extremely difficult to follow.

The second - and longest - section of The Emperor's New Mind is an exposition of the physics that Penrose expects to lie behind the aspects of the mind that would be non-algorithmic: quantum mechanics and, in particular, the resolution of the differences between quantum mechanics and relativity. (This is one of the most important and most difficult problems in modern theoretical physics.) He postulates that the way this incompatibility will be overcome will involve the incorporation of an asymmetric time element into the theory, and this requires a discussion of thermodynamics and entropy. (The question of why time appears to flow and many events seem irreversible when the major physical theories behind them are indifferent to the direction in which time flows is another important problem in today's physics.)

It is in this section, dealing admittedly with three of the most notoriously difficult parts of physics, that the limitations of Penrose's ability to write for the layman become apparent. His explanations seem to simplify in the wrong places, and complex mathematics is presented without sufficient textual exposition to make it palatable. He has not a great deal of space, but does tend to digress to explore areas of the theory that he finds interesting, but which are not relevant to his main theme. That would be fine in a book about quantum mechanics, for instance, but is misplaced when quantum mechanics is being introduced for another purpose.

Following a brief description of the physical construction of the human brain and some experiments to do with consciousness, Penrose rounds off the book with his explanation of how the brain might work, contrasted with the ideas of strong AI enthusiasts. He is not in fact particularly concerned to detail reasons for finding strong AI unlikely, but is more interested with coming up with a physical description of consciousness which would be intrinsically non-algorithmic in parts. He does list some arguments against strong AI, including a most interesting couple of pages about mathematical thought, often considered to be one of the most strongly algorithmic modes of consciousness (because mathematical writing uses a style which can make proofs look rather like algorithms).

The level of detail into which Penrose goes when explaining quantum mechanics and general relativity is unnecessary, and he would perhaps have been well advised to cut this section of the book considerably, expanding the final discussion about intelligence and consciousness instead. That discussion is, after all the main point that Penrose wants to make, the reason why he wrote the book. It is also the most interesting and lucid section. If you already know something about relativity and quantum mechanics, the rest of the book may well seem fascinating, but it cannot be recommended as an introduction to either subject.

If I have been rather critical of Penrose's writing, it is not because I disagree with his position on strong AI, but because I feel that his style is not suited to a popular approach. (I would tend to think strong AI unlikely on intuitive grounds - the way I think and feel doesn't seem mechanically determined to myself - and therefore I would say that the burden of proof lies with its proponents.)

No comments: