How to Build a Mind: Toward Machines with Imagination (review)

[Another review of this book, by Robert Pepperell, follows this review - Ed.]

There are some books that tell us more than we want to know about the author and his or her investigations. This is one of them. It is irritating to be told, quite literally (p. 9), that the message of the book is not comprehensible without tracing the author's "journey," including a rough and ready review of 2,000 years of Western philosophy (in just under 90 pages—whew!), his various jobs, job interviews and people he has met along the way.

One excuse given for this is that the subject of the book—whether machines could be said to have consciousness, or imagination—is more a question of losing our prejudices than advances in technology. Professor Aleksander (of Imperial College in London) takes nearly half the book to trace some thinking about consciousness, from Aristotle to Wittgenstein, to reveal these prejudices. He does so with ginned-up conversations with these folks and later with a ginned-up radio interview with some of our contemporary experts on consciousness, such as Daniel Dennett and Roger Penrose. The conceit could be enough to awaken readers' interests in these authors, but is too sketchy to be useful here.

Aleksander has an underlying pedagogical interest. His unremittingly self-centered meanderings are in aid of the thesis—revealed at the end of the book, a "denouement" to a "detective story"—that consciousness is nothing more than ego-centered world representation. This is exemplified by the emergent properties of neural nets, which code the relationship of the perceiving thing to the world around it: that is not just a cup, but also what I might do with the cup, where it is in relationship to my physical body, what I might drink from it and so on. In this sense, imagination and consciousness are sides of the same coin. Consciousness is simply the representation of facts about the entity itself in the context of its physical world.

It is not much of a detective story, and the spoiler here will not reduce the already most moderate impact of the book. There is no sustained discussion of "emergent" properties and how those differ from ordinary abstraction and short-hand (for instance, a "triangular" cloud, a "dangerous" crowd, a "valuable" collection of stamps). Aleksander gives examples of visually ego-oriented representations (machine input here, other things over there), but does not provide a compelling argument that other aspects of mental processes that, in toto, contribute to the sense of self could be handled in the same way. This is because Aleksander never argues his assumption that (1) the relationship between visual (or other sensory) perceivers and the objects of perception are similar in the relevant ways to (2) the relationship between mental entities and the objects of their desire, fantasy, fear, memory, fibbing and other story-tellings, and so on. Aleksander might be right in his assumptions that these latter relationships, tokens of consciousness, are not materially different from visual relationships, but we do not know that yet. It is a frustrating book, because the author has in fact done a lot of work with neural nets and artificial intelligence, but with the exception of a brief, very general description of an experiment some decades ago, little of his work appears here.

Nothing much works in this book: the conceit of its structure is tedious and will convince no one that the author has hit on the relevant definition of consciousness. The discussion of past and present thinkers is thin gruel, not enough for readers who have read them, and irritating to those with some background. There is no "how to" in this, despite the title, just gently vague ruminations. Sum, ergo sum. It is a book without an audience, save the reviewer.