Beyond Single‐Level Accounts: The Role of Cognitive Architectures in Cognitive Scientific Explanation

Abstract

Abstract:We consider approaches to explanation within the cognitive sciences that begin with Marr's computational level (e.g., purely Bayesian accounts of cognitive phenomena) or Marr's implementational level (e.g., reductionist accounts of cognitive phenomena based only on neural level evidence) and argue that each is subject to fundamental limitations which impair their ability to provide adequate explanations of cognitive phenomena. For this reason, it is argued, explanation cannot proceed at either level without tight coupling to the algorithmic and representation level.Even at this level, however, we argue that additional constraints relating to the decomposition of the cognitive system into a set of interacting subfunctions (i.e., a cognitive architecture) are required. Integrated cognitive architectures that permit abstract specification of the functions of components and that make contact with the neural level provide a powerful bridge for linking the algorithmic and representational level to both the computational level and the implementational level. 2 Explanation within cognitive science has frequently been argued to require multiple domains or levels, and several distinct multi-level accounts of cognitive scientific explanation have been proposed (e.g., Chomsky, 1965;Cummins, 1983;Marr, 1982;Newell, 1982). The account of Marr, however, has received perhaps the most attention. A central tenet of Marr's analysis is that a complete explanation of a device's behaviour requires an account of that behaviour at what he terms the Computational Level (CL), the Algorithmic and Representational level (ARL), and the Implementation Level (IL).While Marr argued that an account at each of the three levels was required for a complete explanation, he also emphasised the primacy of the most abstract of his levels, the Computational Level.Specifically, he argued that while "algorithms and mechanisms are empirically more accessible, […] the level of computational theory […] is critically important from an information-processing point of view [… because …] the nature of the computations that underlie perception [and, by extension, cognition] depends more upon the computational problems that have to be solved than upon the particular hardware in which their solutions are implemented" (Marr, 1982, p. 27). This position has face validity, particularly with respect to naturally intelligent systems where the particular hardware in which the solutions are implemented is shaped by evolutionary factors, and so will arguably be wellsuited, or perhaps even optimised, for the computational problems that an evolved agent such as ourselves must solve (though see Jacob, 1977).Marr was critical of approaches to the understanding of cognitive systems that were not rooted in the CL. For example, in discussing the work of Newell and Simon (1972) on the application of production systems to problem-solving, he wrote that "mechanism-based approaches are genuinely dangerous.The problem is that the goal of such studies is mimicry rather tha...

Interaction in Spoken Word Recognition Models: Feedback Helps

“…Encapsulation of computational and/or algorithmic theory is problematic for several reasons ( Cooper and Peebles, 2015 ). For example, if neural evidence is irrelevant to computational-level accounts, behavioral evidence may also be irrelevant ( Griffiths et al, 2008 ).…”

Big data and the industrialization of neuroscience: A safe roadmap for understanding the brain?

“…A second difficulty comes from Marr's "multiple realizability" argument, which states that the same function can be achieved through any number of different substrates (30,54,61). The impossibility of mapping behavior or function in a unequivocal way on the parametric state of the synaptic or conductance ensemble (defining observed dynamics of the neural net under study) was reproduced in simulation models of Aplysia (62, 63) and vertebrate cerebellum (64).…”

Studying strategies and types of players: experiments, logics and cognitive models

“…As to computational cognitive modeling, Cooper and Peebles (2015) argue that computational cognitive architectures such as ACT-R through their theoretical commitments constrain declarative and procedural learning, thereby constraining both the functions that can be computed (the computational level) and the way that they can be computed (the algorithmic level).…”