A preview of the PDF is not available
Physical Computation and Cognitive Science, by Nir Fresco
Fresco, Nir, Physical Computation And Cognitive Science. Berlin Heidelberg: Springer, 2014, Studies in Applied Philosophy, Epistemology and Rational Ethics, Vol. 12, XXII, 229, 83,29 €. According to the author, the objective of this book is to establish a clearer understanding of computation in cognitive science, and to argue for the fundamental role of concrete (physical) computation, for cognition. He succeeds in both. At the same time he is searching for the adequate scope of computation, repudiating attempts of Putnam, Searle and others, who argued against (classical) computationalism in cognitive science, thereby trivializing computation. The book identifies ambiguities in present day approaches to computation and presents and compares different concepts of computation and their applicability to cognitive systems. The main claim is that for computation to be effective in a cognitive system, computation must be physical (concrete). That requirement is motivated by the development of cognitive theories in the direction of embodiment and embeddedness. The corollary is that the Turing model of computation does not suffice to cover all kinds of cognitive computational processes, as it is a model of a logical procedure describing computation of a mathematical function, while cognitive processes in an organism cover a much broader range of information processing. Fresco presents the computation as a concept that philosophers of computing and computer scientists as well as cognitive scientists understand in multiple ways. He lists seven diﬀerent conceptions of computation predominant in the literature. The argument shows – what should be obvious in any case – that one accepted formalization of a concept (Turing machine) neither precludes reﬂection on its meaning, nor prevents other quite diﬀerent formalizations. At present it is common to approach cognition through computation in a particular formalization based on Turing model of computation. However, computing is much broader than its logical aspects and its physical implementation (dependent also on types of objects manipulated and time-dependent processes of execution) while it is an aspect very central for understanding of cognition. In the same way as the model of informational universe (always relative to an agent) is not trivial because of layered architecture of the informational universe organized in hierarchical structure of levels of abstraction (Floridi 2009) – the dynamics of that informational universe (which is also a computational universe) is not trivial either. But then, as Fresco rightly emphasizes, it is necessary to generalize Turing model of computation to “concrete” (physical) computation. Fresco explores specifically digital physical computation (and he does not insist on a distinction between digital and discrete) – so he deliberately limits his domain. Floridi convincingly argued against digital ontology on the principal grounds (Floridi 2009). Nevertheless, when it comes to practical physical implementation of computation, current digital computers are successfully used for calculation of continua such as found in fluid dynamics. But that is on the modeling side, and the question is only how fine-grained model is sufficient to represent continuous system. The distinction continuous/discrete is not only the property of the physical world; it is a property of the relation between the cognizing agent and the world. (Dodig-Crnkovic and Müller 2009) p. 164. As the basis of an IP (information processing) account of computation, Fresco have chosen instructional information, “prescriptive in that its processing function is aimed at making something happen.” p. 140. The book presents key requirements for a physical system to perform nontrivial digital computation in terms of information processing. The system must have the capacity to: 1. Send information. 2. Receive information. 3. Store and retrieve information. 4. Process information. 5. Actualize control information. (Implementing this requirement is what separates trivial from nontrivial computation.) In the above list of requirements, strong influence of conventional computers is visible. As a summary, on p. 205 Fig. 8.1, there is a diagram showing the relations among the six different accounts of computation analyzed: 6. The most specific account: PSS (physical symbol system) account. UTMs (Universal Turing Machines) and universal stored program digital computers. 7. FSM (formal symbol manipulation) account: program controlled digital computing systems – special purpose TMs, special purpose digital computers 8. Algorithm execution account: digital systems acting in accordance with an algorithm, FSA (Finite State Automata), Hypercomputers 9. The Mechanic and IIP (Instructional Information Processing) accounts: Logic gates, Logic circuits, Discrete connectionist networks 10. The most general account: The “Triviality” account: every physical object computes - Searle-triviality thesis and the Putnam-triviality theorem imply that every sufficiently complex system performs every digital computation In the list above, between items 4 and 5, the account of computing nature is missing, that is the claim that the whole of nature computes, in general as a network of computational networks on different levels of organization (Dodig-Crnkovic 2014). It continuously performs information processing that computes its next state, (Chaitin 2007), where every physical system performs some computation. It is very important to make a distinction between Computing Nature (Stepney et al. 2006; Stepney 2008) (Dodig-Crnkovic and Müller 2009)(Rozenberg et al. 2012)(Zenil 2012) and “Triviality account” in which every physical system performs every kind of computation. Fresco seems to be skeptical about the computing nature approach, as his focus in this book is to present the state of the art and to clear existing muddles around computation and cognition, and not so much to introduce new developments in the field. “It remains an open question though whether embodied computation is indeed the type of computation that takes place in nature. But at least traditionally, it has not been invoked as the basis of the Computational Theory of Mind (CTM).” (p. 4). See also (Fresco and Staines 2014). Even though the CTM does not assume natural computation as a basis of computational approaches to cognition, in the computing nature approach, embodied computation comes naturally from the basic assumptions. If cognition is explained computationally, that computation must be embodied. The fact that traditional CTM did not realize the importance of embodiment points out CTM’s historical limitations. At the time classical computational theory of mind was developed, the belief prevailed that it would be possible to grow and sustain a conscious “brain-in-a-vat”. However, understanding of cognition has increased dramatically since the days of classical CTM, and any respectable contemporary theory of cognition must address embodiment. Fresco in this book makes an important and correct argument that the explanatory frameworks of computationalism, connectionism and dynamicism, contrary to frequent claims are not mutually exclusive but rather complementary approaches, suggesting the way for their integration. Some open questions that remain outside of the scope of the book Physical Computation And Cognitive Science are still of interest and should be mentioned. One fundamental perspective that is missing when it comes to cognition is the biological one. Cognition is a fundamentally biological phenomenon and in order to be able to construct cognitive computational artifacts it is important to understand how natural cognition functions, develops, and evolves. (Maturana and Varela 1980) It is hard to address cognitive phenomena without biological perspective. Computing nature approach includes those aspects and makes them integral part of its discourse. As a consequence of the aims and the framework chosen in the book, computers are taken to be the machines we have today, which also brings some assumptions and constraints that are not necessary. Among others the assumption about necessary infallibility of computation that is implicitly taken for granted, for example in the discussion of miscomputation (p. 41). Turing’ s own view of intelligent computing machines with learning capability is different, as he claims: “There are indications however that it is possible to make the machine display intelligence at the risk of its making occasional serious mistakes.” (Turing 1947) as quoted in (Siegelmann 2013). The allowance for cognitive computation making mistakes and even fatal errors might change the arguments and conclusions offered in the book. The next discussion that I find lacking is the role of explicit account of an agent for whom/which a process is computation. In the computing nature approach, with Hewitt model of computation (Hewitt 2012) in the background, agency-based view of cognition becomes visible and obvious. Instead of having one single definition of computation for all levels of organization, we can define computation in the sense of Hewitt model by interactions between computational agents (actors) that exchange information. The prospect of further development of computational accounts of cognition is nicely outlined in the concluding chapter of the book: “Research attention should be directed toward gaining a better understanding of the types of information processed by natural cognitive agents, how they are processed and interact and how such processing throws light on human cognitive architectures. Such research should examine how cognitive agents produce, acquire, extract, analyze and use information in learning, planning and decision making, for example. It can inform contemporary cognitive science by identifying the mechanisms in the human cognitive architecture that are necessary for these information-processing operations.” p. 225 To sum up, the main virtues of this timely and enlightening book are: systematicity and unusual clarity in eliciting key requirements for a physical system to perform concrete digital computation and providing comparison between different existing approaches to cognition. The book shows clearly that computing in general is broader than abstract models of computation, and cognitive science should be based on it. Gordana Dodig-Crnkovic Chalmers Technical University and University of Gothenburg References Chaitin, G. 2007. Epistemology as Information Theory: From Leibniz to Ω, in G. Dodig-Crnkovic, ed. Computation, Information, Cognition – The Nexus and The Liminal. Newcastle UK: Cambridge Scholars: 2–17. Dodig-Crnkovic, G. 2007. Epistemology Naturalized: The Info-Computationalist Approach. APA Newsletter on Philosophy and Computers, 06/2: 9–13. Dodig-Crnkovic, G. 2014. Modeling Life as Cognitive Info-Computation, in Computability in Europe 2014. eds. A. Beckmann, E. Csuhaj-Varjú, and K. Meer, LNCS. Berlin Heidelberg: Springer: 153–163. Dodig-Crnkovic, G. and Mueller, V. 2009. A Dialogue Concerning Two World Systems: Info-Computational vs. Mechanistic, in Information and Computation eds. G. Dodig-Crnkovic and M. Burgin, Singapore: World Scientific Pub Co Inc. Floridi, L. 2009. Against digital ontology. Synthese, 168/1, 151–178. Fresco, N. and Staines, P. 2014. A revised attack on computational ontology. Minds and Machines, 24/1: 101–122. Hewitt, C., 2012. What is computation? Actor Model versus Turing’s Model, in A Computable Universe, Understanding Computation and Exploring Nature As Computation. ed. H. Zenil, World Scientific Publishing Company/Imperial College Press: 159-177. Maturana, H. and Varela, F. 1980. Autopoiesis and cognition: the realization of the living, Dordrecht Holland: D. Reidel Pub. Co. Rozenberg, G., Bäck, T. and Kok, J.N. eds. 2012. Handbook of Natural Computing, Berlin Heidelberg: Springer. Siegelmann, H.T., 2013. Turing on Super-Turing and Adaptivity. Progress in Biophysics and Molecular Biology, 113/1: 117–126. Stepney, S. et al. 2006. Journeys in Non-Classical Computation II: Initial Journeys and Waypoints. Int. J. Parallel Emerg. Distr. Syst., 21, 97–125. Stepney, S. 2008. The neglected pillar of material computation. Physica D: Nonlinear Phenomena, 237/9: 1157–1164. Zenil, H. ed. 2012. A Computable Universe. Understanding Computation and Exploring Nature As Computation. Singapore: World Scientific Publishing Company/Imperial College Press.