The 1992 version of today's £500 microcomputer will, by present standards, be an extremely powerful machine, with 1 million bytes of main memory. By then, most of the educational software marketed for the current breed of microcomputer will appear rather simple-minded.

In fact, authors Tim O'Shea and John Self say little about existing commercial software, but what they do say is uncomplimentary. It is described as being of "poor" quality, based upon impoverished theories of learning, reflecting little programming skill, and standardised on outdated technology. If we deduce from the programs now flooding the educational market that writing software will remain a spare-time activity then we are mistaken, for computer-aided learning (CAL) systems will increasingly be based on artificial intelligence.

The book traces a general trend in the evolution of CAL away from rigid computer-oriented approaches towards sensitive learner-oriented ones. Various lines of development are described and evaluated and their limitations discussed. Thus, although linear programs, branching programs and simulations have merits, they are seen as making an uninspired, marginal contribution to education.

Approaches based upon generative CAL and mathematical models of learning presuppose well-structured subject matters and suitable learning models. The languages Ticcit and Plato misplacedly emphasised technology, and failed to solve the problems of producing worthwhile teaching materials. Only two approaches, problem-solving and dialogue systems, are seen as likely to enhance the educational process radically, mainly because these approaches do not have clear limitations.

Problem-solving regards the computer as a tool for developing students' problem-solving skills. This is the Piagetian approach exemplified by Logo. The authors think it likely that further languages and environments will be developed, and that these will reflect children's cognitive growth. We will certainly hear more about Smalltalk, a set of object-oriented tools, which at present requires considerable computing facilities but which will be fully implemented on our 1992 microcomputer.

Also available will be expert dialogue systems which contain not only subject knowledge but also knowledge about the student being taught and about alternative teaching strategies. These systems will be capable of holding intelligent tutorial dialogues. Within certain domains, such as symbolic integration and bacterial infections of the blood, dialogues that compare well with those of author-language programs have already been achieved. However, the book does not avoid discussing some of the state-of-the-art difficulties with these system.

Clearly, it is unlikely that teachers will write large problem-solving and dialogue systems themselves, and in any case O'Shea and Self suggest that, as far as schools are concerned, computers will not mimic other teaching processes but will more likely be used to provide programming environments and educational games. It could be argued that it is in this area that the authors underestimate the contribution that can be made by the current educational computing scene in inventing and experimenting with different kinds of computer-based learning experiences. Nevertheless, the book presents a readable account of research and ideas in artificial intelligence likely to have considerable influence on future educational computing. It should be added to the list of essential reading for practitioners and researchers in Computer Aided Learning.