Growing Up With Lucy: how to build an android in twenty easy steps

Steve Grand

Weidenfeld & Nicolson, 229pp, [pounds sterling]16.99

After 50 years, the search for artificial intelligence (AI) has gone nowhere. We know a lot more stuff now, but it is far from clear that any of it has anything to do with intelligence. Meanwhile, artificial life (AL), a project simultaneously more and less ambitious than AI, is equally stalled. Certainly you can read the books, go to the conferences and talk to the people, and you will think we are steaming ahead, that handy androids are just around the corner. The truth is that, conceptually, we are still at square one.

The superficial reason for this is complexity. There are a million cells per square millimetre of the human cortex and each one has 10,000 connections. Artificially constructing such a thing is impossible. Notthat there would be any point in doing so. Good Old-Fashioned AI (Gofai), as it is sometimes known, was based on the belief that the brain must be like a computer and could, therefore, simply be wired up. This was rather like the Victorians believing that the brain resembled a steam engine. Contemporary vanity always demands that our most glamorous machines exemplify something more than their mere machinehood. But they don't.

Connectionist theory, involving the exciting idea of neural networks (Data in Star Trek has one of these), came next. Arrays of artificial neurons were created which, it was hoped, would acquire intelligent characteristics and, finally, boot themselves into consciousness. No luck so far. AL took a step back and tried to create the basic forms of life in the hope that they would evolve, as we once did. Again zilch.

The problem, in fact, may not be complexity at all. As Steve Grand points out, the way the brain works may be very simple. All those neurons are remarkably similar and the way they make connections may follow very elementary rules. But our preconceptions have made us blind to all of this. The answer is to stop being a scientist and start being an engineer.

Grand is almost entirely self-taught. He couldn't get on with conventional education and stumbled into computer programming. There, he developed the game Creatures, in which quasi-autonomous beings lived happily inside your PC. It was way ahead of anything developed before or since. He won an OBE, which so startled him that he was encouraged to plough ahead with his plan to build a robot baby named Lucy.

Lucy has an orang-utan head because Grand felt a baby head would be too disturbing. He has to keep killing her and rebuilding her. After three years she can point at a banana and make an "Arp" sound. Not much, you might think. I met Grand at the beginning of this process and he was certainly expecting much more. He takes a lot of Huttonesque swipes at journalists in this book because, apparently, they reported him saying he was building a conscious creature and he now says he didn't. Well, he did. But never mind.

[ILLUSTRATION OMITTED]

Grand's technique is to engineer Lucy. He doesn't conceptualise what consciousness may be; he simply attempts to solve problems. The reason Lucy can only point at a banana--other robots can do much more--is that Grand will not cheat. He wants to engineer the whole process from the bottom up and the top down. Attempt anything less, and you will just end up with gimmicks like Sony's Aibo robot dog. What you will not get is anything like intelligence.

The bulk of this book is a description, one just about accessible to the layman, of the way Grand has approached the problem of Lucy. He admits his thinking, like that of so many others in this field, may be wrong. But wrong or right, he is never less than interesting. He insists, for example, that the brain is above all a prediction machine. It doesn't live in the real world; it lives in a virtual world of its own devising. There is no other way, for example, that we could shoot a bird in flight: the real-world, real-time input-output process would be too slow. Any physical act involves an intended future created out of an inner narrative, such as where the bird will intersect with the shot. The book is full of such examples. Unfortunately, when the writing is not technically dense, it is infuriatingly jaunty. Grand adopts a hammy, bloke-in-a-pub style to lead us into the difficult stuff as well as to wrap the book up with his homespun philosophy. This is unnecessary, it's not how Grand comes across in person and it will not sell any more books. It is just how condescending publishers and editors think you bring in the punters.

But stepping back from such trivia, what we have here is a further bulletin from the battlefront that Grand has constructed for himself. He believes, almost without exception, that everything anybody has ever said about the workings of the brain is wrong. Even more fervently, he believes in the absolute value of empirical, engineering procedures. The way we get to the mind is by trying to do what it does, not by thinking about what it is.

He may fail. Many--if not most--scientists in this area assume, perhaps hope, that he will. But it will be a noble failure, a defiant act of British flag-waving in the face of the mighty rationalist battalions of MIT. Failure at such a project would itself be a kind of triumph, a refutation of the blind faith in our competence and reason that is the prevailing contemporary delusion. Perhaps Grand's discovery that the brain does not live in the real world is even more significant than he realises.

Bryan Appleyard is the author of Brave New Worlds: genetics and the human experience (HarperCollins)

COPYRIGHT 2004 New Statesman, Ltd.
COPYRIGHT 2004 Gale Group