Explaining the Computational Mind

Abstract

A defense of the computational explanation of cognition that relies on mechanistic philosophy of science and advocates for explanatory pluralism. In this book, Marcin Milkowski argues that the mind can be explained computationally because it is itself computational—whether it engages in mental arithmetic, parses natural language, or processes the auditory signals that allow us to experience music. Defending the computational explanation against objections to it—from John Searle and Hilary Putnam in particular—Milkowski writes that computationalism is here to stay but is not what many have taken it to be. It does not, for example, rely on a Cartesian gulf between software and hardware, or mind and brain. Milkowski's mechanistic construal of computation allows him to show that no purely computational explanation of a physical process will ever be complete. Computationalism is only plausible, he argues, if you also accept explanatory pluralism. Milkowski sketches a mechanistic theory of implementation of computation against a background of extant conceptions, describing four dissimilar computational models of cognition. He reviews other philosophical accounts of implementation and computational explanation and defends a notion of representation that is compatible with his mechanistic account and adequate vis à vis the four models discussed earlier. Instead of arguing that there is no computation without representation, he inverts the slogan and shows that there is no representation without computation—but explains that representation goes beyond purely computational considerations. Milkowski's arguments succeed in vindicating computational explanation in a novel way by relying on mechanistic theory of science and interventionist theory of causation.

The Enactive Automaton as a Computing Mechanism

“…In particular, Varela, Thompson, and Rosch took computation to require representation (1991/2016, p. 40), and presumably thought that the automaton they described did not meet this requirement. However, more recent accounts of computation do not require representation (see, e.g., Egan ; Fresco , ; Miłkowski , ; Piccinini , ), and so may be compatible with Varela, Thompson, and Rosch's enactive theory of cognition. In this section we will focus on just one of these theories, Piccinini's mechanistic account, and demonstrate that according to this account the enactive automaton described by Varela, Thompson, and Rosch straightforwardly qualifies as a (non‐representational) computing mechanism.…”

The Development of Size Sequencing Skills: An Empirical and Computational Analysis

“…Computational modeling in cognitive development comes from two distinct traditions in computer science, that of symbolic and subsymbolic information processing (Boden, ; Klahr, ; Miłkowski, ). Symbolic computation emphasized automated logical theorem‐proving and more generally adult problem‐solving in the 1950s, an approach which gave rise to production systems in the 1970s.…”

“…The computational methodology chosen combines aspects of Bayesian cognitive modeling (Lee, ), dynamical systems modeling (Van Geert, ), and the cognitive architectural approach, housed within a procedural simulation program (Yule et al, ). We chose to build our own modeling framework, due to concerns that the number of free parameters and predefined, and possibly hidden, theoretical constructs would have been very large should a ready‐made cognitive architecture (e.g., ACT‐R, SOAR) have been used (Miłkowski, ). Despite this concern, there are nevertheless many parameters in our model relative to the simple parameters defining the behavioral change.…”

“…Young's model of subtraction competence in the child, alongside many other models of cognition, are available via Yule et al (). As noted above, the production systems formalism has been adopted into contemporary cognitive architectures ACT‐R and SOAR (Miłkowski, ), and incorporate various mechanisms for representing learning and plasticity. Anderson () provides theoretical and practical guidance for ACT‐R implementations.…”

Structural representation and the two problems of content

“…The fourth and most controversial condition is “system‐detectable error” (systems dependent on S‐representations must be capable of detecting an error based on an insufficiency in their correspondence to their target). Advocates of system‐detectable error typically take it to be required, for it is only systems that are capable of detecting a mismatch resulting in their own actions and some target, for which error matters for that system (e.g., Gładziejewski, ; Miłkowski, ; following Bickhard, , ). As we shall see below, system‐detectable error strengthens the idea that content adds something of epistemic value when describing the contribution of an S‐representation to the success or failure of a containing system.…”