It may appear rather precocious for a field of study that is less than 50 years old to pride itself on having a "classical" form and "non-classical" variants. Yet this is how those at the cutting edge of artificial intelligence (AI) describe the theoretical diversity that currently characterises their discipline. The "classical" form - which has also been dubbed GOFAI (Good Old Fashioned AI) - subscribed to a rather idealised view of intelligence as a capacity for abstract problem-solving. In the 1980s, however, a growing number of researchers began to promote a different approach, in which more down-to-earth capacities, such as being able to find your way from A to B without bumping into things, were also valued.

One of the pioneers of this new approach to AI was a dynamic Australian called Rodney Brooks. Brooks is now director of the Artificial Intelligence Laboratory at MIT, and his papers are regarded by many as landmarks. Such lofty heights must have seemed rather distant two decades ago, when Brooks first began to question the tenets of GOFAI. At that time, his ideas were regarded as distinctly crazy. When Brooks presented what has since become his most cited paper, at the Second International Symposium of Robotics Research in 1985, the chair of the conference whispered to a col league that "this young man" was surely "throwing away his career".

In the first few chapters of Robot , Brooks recounts the story of his journey from young Turk to leading light, intertwined as it is with the history of the new AI. Brooks has shaped the field in many ways. Not only has he played a part in many technological developments, from the Sojourner robot that Nasa sent to Mars to the robotic doll My Real Baby, but the list of graduate students who have passed through his lab reads like a roll call of the brightest young stars in contemporary AI. These chapters are scientific autobiography at its best, brimming with the excitement of discovery. The rest of the book, however, is something of a mixed bag, combining elements of futurology with dashes of science, philosophy and ethics.

The futurology is, thankfully, somewhat restrained. Wary of the dangers of speculating about the distant future, Brooks sensibly restricts most of his predictions to technological developments that are likely to occur within the next five or 10 years. For many of these machines, there are already working prototypes. Always sensitive to the practical issues, Brooks explains why some of the contraptions so frequently envisioned in science fiction are unlikely to become reality in the near future. Humanoid robots that can push the vacuum cleaner around for you are, unfortunately, a long way off. If carpet-cleaning robots do become widespread, they are much more likely to look like plastic Frisbees that scuttle around under your feet. Ditto for robots that mow the lawn.

What is missing from Brooks's vision of our robotic future is any sense of the social inequalities that such technological developments are bound to exacerbate. The only passage in which he discusses the role of the developing world in his future technotopia presents a horrific vision of a new kind of sweatshop - one in which armies of workers sit at computer terminals, operating the robots that do the housework in affluent American homes.

The scientific sections are also disappointing, but for a different reason. Nobody is better placed than Brooks to explain the main ideas underpinning the new AI, so it is a real shame to find him repeatedly shying away from the technical details, which are reserved for a short appendix.

Those who wish to see Brooks at his best would be better off buying his previous book, Cambrian Intelligence , which contains eight of his most brilliant essays. These essays were not written for a popular audience, and are all the better for that. However, despite the uneven quality of Robot , it is an enjoyable read, and worth buying for the first few chapters alone, which give a wonderfully personal insight into the remarkable paradigm shift that AI is still undergoing.

Dylan Evans is research officer in evolutionary robotics at Bath University. www.dylan.org.uk