Getting your ACT together

What is happening in the human head to produce human cognition? This question, says John Anderson, drives the work of most cognitive psychologists. The particular model of the mind that Anderson proposes, ACT-R, is a form of production system. Such models have two components, a set of conditional rules, known as productions, and a representation of the current state of the system's "mind''. The state activates those rules whose conditions are satisfied, and the system follows the active rules, changing its state and so making other rules applicable. In this way, a relatively simple system may, with appropriate rules, exhibit complex behaviour, such as reasoning or problem-solving.

Production systems have been around since the late 1960s. There are now many variations on the basic theme. Is the current contribution novel? To answer this, one must appreciate that this work is merely the latest instalment in a story that has been evolving and developing over the past 18 years. The distinctive characteristic of Anderson's research programme is that he associates an activation value with each production (hence the ACT series of production systems), and it is this value that determines the likelihood that any particular production rule will be followed. The added ingredient this time around is what Anderson terms "the rational analysis of the adaptive behaviour of human thought" (hence the "R" for rational in ACT-R). Anderson believes that cognitive skills such as memory, categorisation, inference and problem-solving maximise achievement of information-processing goals while minimising computational costs. This assumption has served to guide the development of the ACT production system over the past five years.

It is in the context of the acquisition of cognitive skills, such as secondary school mathematics, that much of Anderson's work is conducted, and several intelligent tutoring systems have been developed to test and refine the theory. These systems, which have been used on an experimental basis in schools and colleges to teach geometry, algebra, calculus and computer programming, exploit the assumption that human cognition is based on a production system and that learning involves both the acquisition of new and relevant rules and the learning of when to apply those rules. The results of some of these experiments are impressive, with a reduction in learning times of two-thirds in the best cases. Learning tends to be a thorn in the side of traditional computer models of the mind, so the achievements of ACT-R in guiding the development of intelligent tutoring systems should not be taken lightly. However, it remains to be seen whether such systems will become widely used. Anderson argues that many students in real classrooms have little intention of learning, and in such circumstances a change in attitude will produce better results than any tutoring system.

Returning to the question of novelty, it must be said that this book does not inspire. Much of the work is a direct reworking of Anderson's previous research in the context of rational analysis. While rational analysis may be a powerful tool, this reworking does not require 320 pages.

The book comes with a floppy disk containing various bits of software, including the ACT-R production system and many of the examples used throughout the text. An appendix detailing the assumptions of the theory and their relation to the assumptions of previous theory in the ACT series is also included. All these are indicative of the importance that Anderson attaches to methodological concerns. If the research programme is to be progressive it is essential that each version of the theory be publicly accessible and stated in precise terms.

Unfortunately, the disk does not achieve its full potential. Indeed, it is virtually useless without access to a Macintosh computer running the appropriate version of LISP.

Despite this flaw, the book is reasonably accessible. It should be of interest to students and researchers in cognitive psychology and cognitive science, as well as, perhaps, to educationalists concerned with optimising teaching strategies.

Richard Cooper is a research fellow in psychology, University College, London.