Review of:
Sutton, John. (1998). Philosophy and
Memory Traces: Descartes to Connectionism.
Christopher D. Green
Department of
For about the last 15 years, cognitive science has been embroiled in a sometimes-bitter debate about the nature of mental representation. Very roughly speaking, the traditional view is that cognitive processing is the serial manipulation of physical "symbols," each of which locally represents a particular concept or other unit of propositional attitudes. As a result, this position variously goes by the names of symbol(ic)ism, serialism, and localism, among others. The newer view is that cognition involves parallel computation over multiple interconnected "nodes" that individually are semantically uninterpretable, but that collectively constitute distributed mental representations. Such networks go by the names of parallel distributed processing (PDP), connectionism, and neural networks. In the mid-1980s, Patricia Churchland was one of the first philosophers to embrace connectionist cognitive science, and it became closely connected with her Neurophilosophy (1986, MIT Press), which essentially is the view that philosophers of mind should be keeping in much closer contact with advances in the neurosciences if they hope to gain any real insight into cognitive goings-on.
John Sutton's Philosophy and Memory Traces: Descartes to Connectionism appears to be a kind of historical defense of Churchlandian neurophilosophy, particularly with respect to the theory of memory. The book is divided into four major sections. The first, entitled "Animal Spirits and memory traces," runs nearly 100 pp., and contains two substantive chapters discussing the history of the theory of animal spirits. The first of these lays in some conceptual background. The second chapter develops an interpretation of Descartes' physiology of the brain in which it is portrayed as a precursor to modern "distributed" models of cognition. The second section, "Inner discipline," consists of seven short chapters, collectively about 100 pp., on various topics that follow what Sutton describes as "the linked fates of animal spirits and of neurophilosophical models of memory through the century after Descartes" (p.117). In the third section, "The phantasmal chaos of association," about 50 pp., Sutton attempts to draw parallels between Jerry Fodor's modern objections to connectionism and historical critiques of associationism. Finally, in the 50 pp. fourth section, the author attacks directly a number of "standard" criticisms of distributed models of memory.
All this is preceded by a general introduction, outlining the author's main aims. It appears to be straightforwardly polemical. Sutton attempts to deploy historical events in such a way as to establish the credibility of neurophilosophy. It is said to have first arisen in the writings of Descartes, but was distorted and suppressed (see esp. pp. 10-11) by its opponents. Only now do we see its return, in the form of distributed connectionist models of cognition, but back too are its old nemeses, hoping to drive it underground once again. One can hardly help but draw the conclusion that Sutton sees this as an epic struggle between forces of light and darkness, but one in which the "Enlightenment" tendencies toward "rationality" and "literality" are (ironically?) cast in black hats, while "construction," "creativity," and a certain kind of exuberant "chaos" play to role of good guys. If this all sounds a little more postmodern than typical discussions of cognitive science, it is. The introductory chapter is peppered with references to "genealogy," "the body," "engagement," "context," "culture," "holism," "counter-theory," and the like. Sutton writes in a literary style, to be sure: "Molly Bloom," he tells us at one point, "not Sherlock Holmes, is the fictional figurehead for the neurophilosopher" (p. 17).
I see two key difficulties here. First, it is not at all clear how connectionism advances a postmodern or neo-Wittgensteinian species of social philosophy any more than traditional symbolic cognitive science does. Even Sutton concedes (p. 5) that the leading neurophilosophers have not had much to say about the social, apart from a few motherhood clauses. Many connectionists are even more reductionistic than their symbolist opponents -- even propositional attitudes are too "airy-fairy" for them. Getting closer to the neural level of description is the aim of most connectionists, not getting social about the mind. Sutton's endorsement of distributed memory models seems to stem from his belief that they "allow for and invite such extensions [to social and ethical domains]" to a greater degree than symbolic models. Second, his characterizations of symbolic memory models are so utterly one-sided that we are hardly left even with straw men. Traditional theories are said to relegate memory to "static records in cold storage" (p. 1), an "archive caricature … [that] must be … lampooned" (p. 4), and "a dusty corner" (p. 6). This kind of hyperbole may have its place, but Sutton makes no attempt to seriously grapple with the work of the best (if "localist") memory researchers of the century: where are Miller, Broadbent, Milner, Baddeley, Craik, Squire, Tulving, etc.? They are "erased" -- to borrow a term from Sutton's philosophical tradition -- in favor of metaphor and allusion.
So, considered as quasi-historical intellectual polemic in the service of contemporary neurophilosophy, how does the rest of the book shape up? Chapter 2 is an interesting but impossibly short romp through the semantic cross-currents of the theory of "animal spirits" prior to Descartes. Relying almost entirely upon secondary sources, Sutton touches on the theological, biological, occult, medical, and moral connotations of "spirit" in pre-modern centuries. Sutton's stated aim is not to revive sprit theory, per se, but rather to show that its multifarious semantics may tell us something important about the way we should conceive of memory today.
Having "softened up" his readers for new perspectives on old, rejected accounts of mind, Sutton attempts in Chapter 3 to reclaim Descartes from present-day rationalists (e.g., Fodor, Chomsky) by re-interpreting his speculative brain physiology as a precursor to present-day distributed models of memory. Sutton repeatedly attempts to distance himself from the anachronistic implications of such a project, but this is difficult to countenance. The chapter only makes sense as an attempt at "discipline building" by appropriating an authoritative historical figure to the side of neurophilosophy. But still, the chapter should not be dismissed. Sutton has important points to make, and his attempt to make sense of Descartes' physiology is welcome Sutton also carefully outlines pre-emptive replies to the most obvious criticisms of his interpretation. There can be little doubt, however, that the motivation at work is more about rhetorical strategy than about gaining insight into Descartes' thought.
The second section of the book highlights English responses to Descartes' memory theory. Chapter 4 parallels the "realism" of 17th-century mechanists with that of modern connectionists such as Paul Churchland (p. 123). Sutton says the two positions "chime well," but one cannot help but warily notice that their respective opponents -- "Aristotelian scholastics" and anti-realists such as Van Fraassen, respectively -- have little in common. Sutton also suggests that Robert Hooke's interest in superposition and interference patterns in the physics of light may have influenced his position on memory. Chapter 5 briefly examines that the reactions of four Englishmen -- Hooke, Kenelm Digby, Jospeh Glanvill, and Henry More -- to the Cartesian theory of memory, with the aim of demonstrating that Descartes' advocacy of distributed representation, rather than his dualism, was the main target of their objections. There are some interesting details here, but the chapter is unhappily short. In Chapter 6, Sutton steps back from his historical narrative to clarify two distinctions on which his argument depends: local vs. distributed mental representation and implicit vs. explicit mental representation. The second of these, in particular, has provoked enormous debate in psychology, not just with respect to connectionist models of cognition, but with respect to memory and learning more generally. Many believe the distinction to be fundamentally incoherent. Sutton does not seem to recognize the connection of his specific issue to the these more general debates, and the import of his comments suffer accordingly. In Chapter 7 he seems to argue that even Locke was "moving towards" a distributed view of memory (and the person), but this view turns on the questionable identification of Locke's description of memories being "revived" ideas with the modern belief that memories are "reconstructed" (p. 168). In the end, Sutton waters his claim down to one in which Locke's own model is admitted to have been "localist," but that he was "aware, too, of the other [memory] phenomena which a distributed model takes seriously" (p. 173). Chapter 8 briefly explores the reasons that spirit theory continued to be popular, even after 17th-century experiments showed that muscles do not increase in volume when contracted (and thus could not have been infused with additional "spirit"). Despite some interesting historical material, the analysis seems to be tainted with Sutton's implicit expectation that scientists -- even of the 17th and 18th centuries -- would feel honor-bound to abide by strict Popperian falsificationism. Naturally enough, they did not, and alternative interpretations of the experiments allowed "spirits" to live another day. The moral implications of spirit theory, which constitute an increasingly significant theme in the early chapters, are brought to a head in Chapter 9, which traces the 18th-century "crisis" in spirit theory that arose from the full realization of its irrationalist implications. Not just the free and rational will, but also the forces of disease, distemper, weakness -- even the Devil -- were thought to affect the distribution of spirits, causing all manner of erroneous perception, belief, memory, and action. The implication seems to be that this threat to Enlightenment ideals did more to drive spirit theory from scientific respectability than any empirical findings or theoretical innovations, but Sutton refrains from asserting this definitively, preferring instead to "consider" five alternative hypotheses about its decline in chapter 10.
Section III opens with a short chapter (11) on Jerry Fodor's criticisms of connectionist cognitive science. Interestingly, the main target of the Sutton's attack is not Fodor's philosophy of mind, but his philosophy of science, which Sutton surprisingly suggests parallels 18th- and 19th-century moral critiques of associationism. In Chapter 12, he tries to show that criticisms of classical associationism do not bear against "neo-associationism" (i.e., connectionism) because, as he puts it at one point, "no things [e.g., ideas, neurons] are associated at all in distributed models" (p. 243). This is supposed to be an advantage, but the reader not committed to connectionism in advance is left wondering what, exactly, such models model if not "things" of any sort. What relation, exactly do the nodes and connections of the networks bear to "natural" cognition? Chapter 13 examines the relation of Hartley's vibrational theory of memory to modern distributed networks, but, as is typical of the book, does so extremely breifly. The final chapter (14) of the section critically skims through Reid's and Coleridge's efforts to refute Hartley and his associationist colleagues.
In the fourth and final section of the book, Sutton shifts from the historical to the contemporary -- "clearing the ground …. by showing that distributed models … do not suffer the conceptual incoherence with which critics charge them" (p. 277). Chapter 15 attempts to employ distributed models of memory to "dissolve" the dispute between "direct realism" (à la J.J. Gibson) and representationalism. The effort is somewhat quixotic, as Gibsonians represent a tiny minority of perceptionists, and virtually no memory researchers. Nor does Gibson appear much in the connectionist literature, though Sutton keenly points up some exceptions. Gibson is presented as a figure who can let Sutton have his neo-Wittgensteinian cake and eat it too (i.e., reject the "storage" memory theory in favor of a "social," "dynamic" but still "scientific" theory). Chapter 16 rehearses several standard criticisms of "trace" memory theories, and argues that distributed models -- particularly "unsupervised" connectionist networks -- do not fall victim to them. In the final chapter (17), Sutton makes a plea for accepting the "chaos" of memory processes rather than "imposing" an artificial "rationalist" order upon them.
Many aspects of this book -- especially the chapters on Descartes and Locke -- are interesting. In the end, however, I found it unconvincing. I wonder exactly which constituency Sutton was writing for. Those from the "localist" tradition are likely to regard it as a transparent attempt to extract historical authority for connectionism from tendentious re-interpretations of past eminent philosophers. The connectionist community might be less critical because he's on their "side," but Sutton's sometimes-florid style and Wittgensteinian sensibilities will not resonate well there. Finally, Wittgensteinian critics of all attempts at cognitive science are not likely to be convinced by Sutton's efforts to set connectionism apart from other modes of "objective" mental modeling. Connectionism does little to address either the normative issues involved, or the (correct) impression that, in the final analysis, connectionist network are every bit as mechanistic as the localist, symbolic models they aspire to displace.