The Routledge Handbook of the Computational Mind
Mark Sprevak and Matteo Colombo
Reviewed by Adrian Downey
The Routledge Handbook of the Computational Mind
Mark Sprevak and Matteo Colombo (eds)
London: Routledge, 2019, £190.00
ISBN 9781138186682
In their introduction to this volume, Mark Sprevak and Matteo Colombo highlight its three aims: to constitute a ‘time-capsule’ of contemporary research of use to future (historical) scholars; to serve as a state-of-the-art resource for contemporary scholars; and to function as a teaching resource. To let the cat out of the bag from the outset, this volume succeeds on all three fronts, and then some. It provides an authoritative and comprehensive summary of virtually all the work carried out within the aegis of the computational theory of mind (henceforth, CTM).
The volume is divided into four parts, which deal, respectively, with the history of the CTM, the different approaches to computation pursued by CTM researchers, contemporary debates and outstanding issues in computational research on the mind, and practical examples of the CTM paradigm in empirical action. Each of its thirty-five chapters—and also the introduction—is succinct, packed full of detail, and easy to digest. Moreover, there is remarkable co-ordination of, and little overlap between, the chapters. This gives the volume a Gestalt quality wherein we get a grasp not only of the individual research topics, but also the broader remit and trajectory of CTM research as a whole. In what follows I will briefly summarize the volume’s chapters, highlight some intriguing core themes of each part, and seek to convey why it constitutes an excellent and informative contribution to the literature in the process.
Part 1 of the volume charts the historical trajectory of the CTM, beginning with a nice summary of its historical antecedents which are, by and large,[1] to be found in the Early Modern period (Isaac, Chapter 1). Next (Proudfoot and Copeland, Chapter 2), we are provided with an account of how the ‘universal Turing machine’ went from abstract idea to being concretely instantiated in physical computers. This account chronicles both the timeline of these developments and the media’s reporting of them. The cybernetics movement of the mid-twentieth century receives two treatments, one focusing upon its development in Britain (Dewhurst, Chapter 3), the other highlighting its interventions in and contributions to the formulation and debate of the CTM (Abraham, Chapter 4). The CTM’s development is also the topic of the next chapter (Aizawa, Chapter 5), which argues that scrutiny of canonical papers in cognitive science—those given at the 1956 ‘Symposium on Information Theory’, often considered a watershed conference for the field—reveal their authors to have been little concerned at that time with a computational understanding of mind. The connectionist paradigm subsequently receives a detailed survey, outlining its pre-history, ‘Golden Age’, and highlighting contemporary issues and debates (Buckner and Garson, Chapter 6). With the historical overview of the CTM provided, Part 1 fittingly ends with a speculative look to the future, focusing in particular upon the important philosophical issues that technological advances in computing may bring to the fore (Shanahan, Chapter 7).
Overall, Part 1 functions as a useful corrective to the ‘Whig history’ of cognitive science one often encounters—that there was a seamless, almost inevitable transition from the development of computers to the CTM characteristic of the ‘cognitive revolution’ in cognitive science—with the chapters making clear that things are not so clear-cut. Especially welcome in this regard is the prominence given to the cyberneticists, who constituted crucial precursors to the CTM and yet are often overlooked in contemporary discussion. Historical accuracy alone requires their mention, and especially given that work in cybernetics pre-figured many contemporary debates about the CTM (and perhaps, as chapters on cybernetics suggests, hold the keys to their resolution), it is nice to see the cyberneticists quite literally accorded their rightful place in the history books. Another interesting theme of Part 1 is the resounding constancy and relevance of philosophical concerns to the CTM throughout its history, a point I shall discuss later in the review.
Part 2 contains detailed exegesis of the various computational approaches toward (modelling) the mind, beginning with an outline and exhibition of classical computational models (Samuels, Chapter 8), which model cognition as the algorithmically governed (and semantically interpretable) manipulation of symbolic entities. Connectionist models are discussed in the next chapter (Chapter 9), and Catherine Stinson argues that modellers in this area are primarily interested in discerning the ‘basic principles’ at work in neural processing and the ‘generic mechanisms’ responsible for instantiating them. A dynamical information processing framework, which incorporates aspects of the putatively competing computational and dynamical accounts of cognition, is next provided (Faries and Chemero, Chapter 10). Following this, we receive an informative overview of probabilistic models, which highlights the insights, strengths, and pitfalls of their various permutations (Danks, Chapter 11). Part 2 concludes with a whistle-stop tour of the prediction error minimization (predictive processing) framework. Jakob Hohwy (Chapter 12) explains why we might think brains are primarily in the business of minimizing divergences between expected and actual input, and what we would expect to see the brain doing if this hypothesis is correct.
What is most notable about Part 2 is the sheer variety of computational frameworks on offer. Given that they differ substantially from one another, it may appear they are engaged in a competitive, winner-takes-all battle. Upon reading through these chapters, however, it quickly becomes apparent that this is not the case (at least, that it is not obviously so). There exist interesting similarities and overlaps between these apparently conflicting approaches. Furthermore, even where there are differences, it often seems that these differences reflect nothing more concrete than the idiosyncratic interests and goals of a given group of researchers. These observations about Part 2 suggest that a perspectivist or pluralist understanding of science—that science proceeds via the selective modelling of specific aspects or parts of real-world systems guided by the historically and socio-culturally permeated aims of researchers—best captures work carried out under the banner of the CTM, as the editors themselves highlight in their introduction to the volume.
Part 3, the largest in the volume, guides us through a variety of aspects of, and issues within, the CTM. Trivialization concerns—that the CTM is trivial because all physical systems can be understood to implement computations—are discussed first. Mark Sprevak (Chapter 13) argues that while trivialization arguments ultimately succeed, it is helpful to try to ward them off, because our attempts to do so birth good accounts of implementation (a plurality of which should be endorsed and applied in a context-sensitive fashion). Implementation is discussed further in the next chapter, with J. Brendan Ritchie and Gualtiero Piccinini (Chapter 14) maintaining that a good account of implementation will respect both psychological and neuroscientific levels of explanation. As such, ‘if the mind is computational in an interesting sense, then psychology and neuroscience directly constrain each other’ (p. 202). There next follows a detailed focus upon levels of explanation, wherein Lotem Elber-Dorozko and Oron Shagrir (Chapter 15) discuss and interpret Marr’s three levels (computational, algorithmic, and implementational), provide reasons to be wary of models that ignore implementational details, and highlight the shortcomings of mechanistic accounts of computation. Next, Daniel Weiskopf (Chapter 16) argues that smooth reductions from psychological models to neuroanatomy never work in practice, and he suggests a non-reductive alternative that allows for realism about psychological models even if they cannot be reduced to the brain. A recounting of ‘Helmholtz’s vision’—the idea that we might provide a physical, computational account of the mind—follows, charting its progress up to the present day, and focusing in particular on the recent advances in and on the shortcomings of brain imaging techniques (Glymour and Sanchez-Romero, Chapter 17).
Frances Egan (Chapter 18) next introduces the key desiderata that a theory of sub-personal representational content should meet. She contends that a deflationary account—which is realist about the mathematical but not cognitive contents posited in models—is best suited to meeting these desiderata. We are then supplied with an informative overview and defence of S-representation (structural, surrogative, and/or simulation representation) accounts that is optimistic about their potential (Ramsey, Chapter 19). The next chapter (Hutto et al. Chapter 20) bucks the trend of the entire volume in expressing outright scepticism about the CTM. The authors argue that representational and non-representational accounts of computation alike face serious problems. They contend that we should radically reformulate our basic understanding of ‘computation’: it is cognition that gives rise to computation, and not the other way around. The practice of neuroscience is the topic of Rosa Cao’s chapter (Chapter 21), wherein she argues that it does not license a representational understanding of the CTM.
Colin Klein’s contribution (Chapter 22) encourages us to embrace arguments sceptical of the ability of the CTM to accommodate consciousness. This is because they help to tighten up our computational accounts and so arrive at ones that can accommodate consciousness (if such a thing is possible). A novel proposal for understanding symbols and concepts within the language of thought/classical computational theory of mind framework follows (Salisbury and Schneider, Chapter 23), after which we are treated to a nuanced discussion of embodied cognition, which concludes that a ‘moderate’ account—one that takes into consideration the insights of both computational and embodied frameworks—is to be preferred (Miłkowski, Chapter 24). Part 3 ends with a demonstration of the usefulness of computational complexity theory for empirical work on the CTM. By employing representative examples, Jakub Szymanik and Rineke Verbrugge (Chapter 25) demonstrate how thinking about computational tractability can help us study and understand minded capacities.
A recurrent theme throughout Part 3 is the importance of what may first appear to be esoteric philosophical issues. Questions like ‘Can a backpack compute minded or even conscious functions?’, ‘What is a computational representation?’, and ‘How can we track the inter-relation between different scientific “levels” of explanation?’ seem precisely the sorts of armchair, irrelevant, and impractical musings that give philosophers a bad name. However, the chapters in Part 3 time and again provide evidence that thinking about and addressing precisely these concerns is crucial for developing an empirically robust account of the mind. Aside from performing the negative role of pinpointing absurdities and weak points in current empirical work, they guide us to better reformulations of our theses and concepts, which, ultimately, improves our understanding of it. Advances in computing, rather than obviating the need for philosophy, appear only to make philosophical issues more relevant and concrete. Philosophy thus looks to be alive and well within the CTM, with philosophical and empirical endeavours being mutually supportive of one another.
Part 4 showcases the plethora of practical research carried out under the CTM banner. Computational cognitive neuroscience is first up, with Carlos Zednik (Chapter 26) critically assessing the (de)merits of ‘top down’ and ‘bottom up’ approaches and concluding we should adopt a ‘bi-directional approach’ that encompasses aspects of each. The next chapter (Irvine, Chapter 27) focuses specifically on the role of simulations within this field and provides a spirited defence of ‘top down’ approaches. We are then given an overview of computational accounts of learning and reasoning, and it is argued that work in this area supports an ‘enlightened empiricism’ about learning and puts pressure on the idea that we possess functionally distinct reasoning systems (Colombo, Chapter 28). Turning to vision, Mazviita Chirimuuta (Chapter 29) explores the extent to which vision is literally (rather than metaphorically) computational. She concludes that a computational approach is one of many fruitful stances we can adopt toward the study of vision, and that work in this area is best understood in pluralist and perspectivist terms. Nico Orlandi (Chapter 30) then argues that disjunctivist and ecological accounts of perception can be accommodated by the CTM provided computation is conceptualized as non-representational. In the next chapter, Micheal Rescorla (Chapter 31) provides an accessible and yet richly detailed overview and defence of (Bayesian) optimal feedback control models of motor activity, and this is followed by an excellent and easy to understand summary of the basics of emotion research and computational approaches toward it (Gu, Chapter 32). Computational psychiatry is then elucidated via example, the authors (Brugger and Broome, Chapter 33) focusing upon reinforcement learning and predictive processing models of schizophrenia. Turning now to the social domain, the penultimate chapter (Michael and MacLeod, Chapter 34) focuses upon the potential for predictive processing to provide a unificatory framework for the study of social cognition, while the final chapter (Huebner and Jebari, Chapter 35) surveys a gamut of group-level activity—ranging from co-ordination between groups of individuals to (arguably) full-blown group-level agency—from a computational perspective.
It is not uncommon for human beings to compare the mind to the pre-eminent technology of their era. Critics of the CTM often contend that it is a faddish product of its time, the latest in a long line of doomed attempts to equate the mind with our latest creations. The sheer empirical depth and detail presented in Part 4—which I simply do not have the space to adequately convey—serves as a strong rejoinder to this idea. Reading through the chapters, one cannot help but be struck by the enormity and scope of empirical work carried out across the full range of aspects and facets of the mind, under the aegis of the CTM. Indeed, it quickly becomes clear that the CTM constitutes a serious, rigorous, and fecund programme of research. This does not mean that the CTM is faultless or beyond rebuke, but it shows that disputing the paradigm requires more than off-hand references to failed analogies of the past. Critics must grapple with the theoretical maturity, empirical scope, and comprehensiveness of the CTM, while, ideally, supplying alternative visions that rival it in these areas. It is presently difficult to see any rivalrous contenders up to the challenge.
Another interesting theme in Part 4, which I again lack the space to fully discuss, is the pervasiveness of the predictive processing approach, with all but one of the chapters discussing it (to varying degrees of detail). There appears to be rampant optimism within the field that predictive processing will do for the mind sciences what Darwin did for biology: provide the over-arching theory required to unite the disparate strands of work within one over-arching (computational) framework. While this optimism is nevertheless tempered by empirical caution—even the most enthusiastic proponents of predictive processing grant that there are many open questions surrounding the framework—one’s subjective impression that predictive processing constitutes the ‘next big thing’ in cognitive science would appear to be vindicated by the work surveyed and discussed in Part 4.
The CTM, whether you like it or loathe it, is presently the most popular and strongly entrenched research paradigm in both the philosophy and science of mind. Consequently, anyone who wishes to understand research in this area must have a firm grasp of computational proposals about the nature of mind. I can think of no better or more efficient way to achieve said understanding than reading this volume and studying it carefully. In providing an excellent synopsis of the history, theoretical understandings, cutting-edge issues, and empirical research programmes of the CTM, this volume constitutes a comprehensive and rounded explication of everything you need to know about the computational research paradigm. I personally benefited hugely from grappling with this volume, as will any reader, whether an interested novice or an advanced researcher.
Acknowledgements
I gratefully acknowledge the help of Elmarie Venter and Alfredo Vernazzani, both of whom provided very useful feedback on an earlier draft of this review. The writing of this review was financially supported by the Alexander von Humboldt Stiftung in the form of a postdoctoral fellowship.
Adrian Downey
Ruhr Universität Bochum
A.Downey@sussex.ac.uk
Notes
[1] Some theorists and ideas discussed are from the period of time shortly after the Early Modern period.