Head jelly that slips down easily

Books about possible neuro-scientific explanations of mental operations seem to be appearing rather frequently these days. The subject matter is bound to be - or at least deserves to be - complex, but the books often can be divided into "easy/difficult" and "difficult/easy" reads. The latter applies, say, to some current books on brain mechanisms of consciousness, where the substance is assumed to be relatively tractable and self-evident, but the analysis is not.

This book falls in the former category. It is an easy read, touching on an extraordinarily wide range of difficult detailed neuroscientific and artificial intelligence research on memory in a style that is fluent, engaging, and readable, and shows the author to be generous and erudite. Its difficulty flows from the very diversity of material and, perhaps even more so, from its mixing of levels of discourse and its assumption that nettles can be rendered painless with verbal analgesics.

Joaquin Fuster is a well-known, pioneering neurophysiologist who for several years has been recording from the brains of animals while they are performing a particular type of memory task. Because the electrical responses of nerve cells are best studied over a limited time interval, Fuster has concentrated on what is often called "short-term delay" tasks, in which an animal must hold perceptual information for some seconds in order to make the correct response. During the initial presentation, the subsequent delay, and the final response, neurones can be found that are responsive during the delay period and can be shown to be specific to the original stimulus.

From such a relatively narrow research front, he argues against what has been a dominant position in memory research for the past two decades. Evidence from humans suggests the brain has a multiplicity of "memory" systems. Patients who cannot recognise or recall events after a minute or two nevertheless demonstrate excellent ability to learn and remember if tested by "implicit" and indirect methods.

A different pathology, in contrast, impedes the learning of motor skills or of "habits". And different lesions selectively impair "short-term" memory (eg, reciting a telephone number back after 30 seconds) without interfering with the ability to memorise lists of items. And so forth. Also some nerve cells have been shown to be remarkably specific in their tuning - some responding to faces, but not to hands, and others just the opposite.

Fuster wants none of that. All these diverse forms are examples of large brain circuits that underlie all manifestations of memory. In this sense, he is arguing for a return to the old notion of "mass action" - that more or less the whole brain was involved in learning and memory, with no more specialisation than the molecules in a bowl of jelly - which was effectively discredited by animal research and human clinical material alike. But it is not pure regression, for he joins hands with the new generation of "neural network" theorists for whom distributed systems are a useful theoretical assumption. He also gives examples of how widely separated regions of the brain may respond in a disparate set of tasks. But to smash a putative kernel of neural dissociationism and specificity requires a heftier sledge-hammer than that wielded here.

He does allow, and presses for, one distinction, namely between perceptual and motor memory, and because the evidence for perceptual systems is already relatively familiar, one value of the book is the weight that it gives to motor learning. But beyond that he assumes that many of the familiar distinctions in the study of memory can be dissolved by immersing them in a suspension of complex, widely dispersed associative networks.

To take one example, he disputes the distinction between short and long-term memory, even though the single unit studies he cites involve delays of the order of only 15 seconds, much too short to be convincing. Surprisingly he asserts there "is no convincing evidence from cortical lesion studies. . . to support the concept of two memory stores (ie, for short and long-term), for vision or any other modality."

One cannot make evidence from Baddeley and Warrington, to cite just one well-known study among several, disappear by such a breath-taking assertion. There are sources, together with cogent theorists, who have argued for several qualitatively different (but normally interacting) memory systems in the brain.

Similarly, he dismisses the evidence for dissociations between category-specific deficits in semantic memory because "close scrutiny often reveals that categorical differences may be attributable to spurious factors. . . Double dissociations, by lesion and category, have not been demonstrated persuasively." The area is far from free of controversy, but also far from devoid of striking examples of double dissociations.

One can acknowledge that theoretical connectionistic networks can be demonstrated to have considerable power to account for heretofore unassailable "higher" faculties, and Fuster's ideological stance is frequently displayed. But he blandly asserts that all problems of memory are basically associationistic. Beyond that, "all mnemonic rules and strategies are based on this assumption and amply validate it". For rule-governed learning to be reduced to just another example of association would indeed be remarkable, but frustratingly the reduction is not enriched by any details of argument.

Curiously, as a thorough-going associationist he is reluctant to embrace a thorough-going materialistic account, and includes a final chapter on "phenomenology of memory" because "it is practically impossible to discuss the dynamics of primate memory without reference to conscious experience". "Cognitive science without consciousness is bound to be, if not sterile, of little relevance to human memory."

As regards consciousness, his position is a familiar one in network theory, namely that it is merely a higher level of spreading excitation, which exceeds a "C- (consciousness) threshold", whereas processing per se need only reach a lower "P- (processing) threshold." But what is the value of exceeding the C-threshold? At times throughout the book he reveals a lingering folksy dualism, eg, "novel programs and plans are formed and stored by the organism in the prefrontal cortex." "In its active state . . . a memory network controls behaviour and presumably gains access to consciousness." (italics added).

But these comments should not detract from what is the attractiveness of the book when used as an entree into a rich fabric of evidence. He provides ready and up-to-date access to a variety of theoretical approaches and methodologies in the study of memory and other areas of cognition, extending even from the theories of the economist Hayek to the representations of Von der Malsburg, the re-entry ideas of Edelman, to evidence from clinical neuropsychology, to a wealth of evidence from single neurone studies, including his own, as they bear not only on memory but also on such issues as perceptual binding.

Readers will find much to whet their appetites even though they may have to go elsewhere for mastication and digestion.

Lawrence Weiskrantz is emeritus professor of psychology, University of Oxford.