Neural Representation

Neural representations are models of the organism and environment built by the nervous system. This chapter provides an account of representational role and content for both indicative and imperative representations. It also argues that, contrary to a mainstream assumption, representations are not merely theoretical posits. Instead, neural representations are observable and are routinely observed and manipulated by experimental neuroscientists in their laboratories. If a type of entity is observable or manipulable, then it exists. Therefore, neural representations are as real as neurons, action potentials, or any other experimentally established entities in our ontology.

The First Computational Theory of Cognition

McCulloch and Pitts were the first to use and Alan Turing’s notion of computation to understand neural, and thus cognitive, activity. McCulloch and Pitts’s contributions included (i) a formalism whose refinement and generalization led to the notion of finite automata, which is an important formalism in computability theory, (ii) a technique that inspired the notion of logic design, which is a fundamental part of modern computer design, (iii) the first use of computation to address the mind–body problem, and (iv) the first modern computational theory of cognition, which posits that neurons are equivalent to logic gates and neural networks are digital circuits.

Mechanistic Functionalism

The first three chapters introduced mechanisms, including functional mechanisms—that is, mechanisms that have teleological functions. This chapter introduces a mechanistic version of functionalism. Functionalism is the view that the nature of something is functional. Mechanistic functionalism embeds this claim in the functions of mechanisms and their components. Mechanistic functions are inseparable from the structures that perform them at the relevant level of organization. Weak (mechanistic) functionalism entails multiple realizability; strong (mechanistic) functionalism entails medium independence. Thus, even if medium independence is closely related to computation, (mechanistic) functionalism about cognition does not entail that cognition is computational. In addition, (mechanistic) functionalism entails neither traditional anti-reductionism nor the autonomy of the special sciences.

Introduction

This book defends a neurocomputational theory of cognition grounded in a mechanistic, functionalist, egalitarian ontology. I argue that biological cognitive capacities are constitutively explained by multilevel neurocognitive mechanisms, which perform neural computations over neural representations. Providing a scientific explanation of cognition requires understanding how neurocognitive mechanisms work. Therefore, the science of cognition ought to include neuroscience to a degree that traditional cognitive science was not expected to. Scientists on the ground have been working on this for a while. Psychology is becoming more and more integrated with neuroscience....

The Church–Turing Fallacy

The Church–Turing thesis (CT) says that, if a function is computable in the intuitive sense, then it is computable by Turing machines. CT has been employed in arguments for the Computational Theory of Cognition (CTC). One argument is that cognitive functions are Turing-computable because all physical processes are Turing-computable. A second argument is that cognitive functions are Turing-computable because cognitive processes are effective in the sense analyzed by Alan Turing. A third argument is that cognitive functions are Turing-computable because Turing-computable functions are the only type of function permitted by a mechanistic psychology. This chapter scrutinizes these arguments and argues that they are unsound. Although CT does not support CTC, it is not irrelevant to it. By eliminating misunderstandings about the relationship between CT and CTC, we deepen our appreciation of CTC as an empirical hypothesis.

Functional Mechanisms

This chapter articulates a goal-contribution account of teleological functions. Teleological functions are causal roles that make a regular contribution to the goals of organisms. Goals can be biological or nonbiological. Biological goals are survival, development, reproduction, and helping. Nonbiological goals are any other goals pursued by organisms. Appropriate situations for the performance of a function are situations in which performing a function provides a regular contribution to a goal of an organism, unless there are more urgent functions to perform. Appropriate rates at which functions should be performed are rates that provide adequate contributions to the goals of an organism, unless there are more urgent functions to perform. Aside from the constraints imposed by tradeoffs between different functions, any condition that lowers the performance of a function below its adequate rate of performance in an appropriate situation results in malfunction.

The Cognitive Neuroscience Revolution

This chapter outlines a framework of multilevel neurocognitive mechanisms that incorporates neural representation and neural computation. Paradigmatic explanations in cognitive neuroscience fit this framework and thus cognitive neuroscience constitutes a break from traditional cognitive science. Whereas traditional cognitive scientific explanations were supposed to be distinct and autonomous from mechanistic explanations, neurocognitive explanations are mechanistic through and through. Neurocognitive explanations aim to integrate computational and representational functions and structures across multiple levels of organization in order to explain cognition. To a large extent, practicing cognitive neuroscientists have already accepted this shift, but philosophical theory has not fully acknowledged and appreciated its significance. As a result, the explanatory framework underlying cognitive neuroscience has remained largely implicit. This chapter explicates this framework and demonstrates its contrast with previous approaches.

Mechanistic Models of Cognition

Psychological and neuroscientific explanations strongly constrain one another, so much so that psychology has become an integral part of cognitive neuroscience. The functional analyses of classical cognitive psychology can be integrated with neuroscientific explanations to form multilevel mechanistic explanations of cognition. At each level of mechanistic organization, a mechanism explains phenomena by showing that they are produced by suitably organized components. This requires abstraction from irrelevant causes and lower level details, which abstraction is an essential aspect of mechanistic explanation. Therefore, psychological and neuroscientific explanations are not autonomous from one another.

Computation and Information Processing

This chapter presents a mechanistic account of physical computation and elucidates the relation between computation and information processing. Physical computation is the processing of medium-independent vehicles by a functional mechanism in accordance with a rule. Physical computation may be digital, analog, or of other kinds. Individuating computational vehicles and the functions a system computes requires considering the interaction between a system and its immediate environment; in this sense, computational individuation is externalistic. Information processing is the processing, by a functional mechanism, of vehicles that carry information. In general, computation can occur without information processing and information processing can occur without computation. Nevertheless, typical computing systems process information, and many information processors are computing systems.

Mechanisms, Multiple Realizability, and Medium Independence

This chapter provides an account of realization within a mechanistic framework and introduces the notions of variable realizability, multiple realizability, and medium independence. Realization is the relation between a higher-level property and the lower-level properties of which it is an aspect. Variable realizability occurs when the same higher-level property can be realized by different lower-level properties—different lower-level properties share the same aspect. Variable realizability is ubiquitous yet insufficient for multiple realizability proper. Multiple realizability proper occurs when the same higher-level property can be realized by different lower-level properties that constitute different mechanisms for that property at the immediately lower mechanistic level. Medium independence is an even stronger condition than multiple realizability: it occurs when not only is a higher-level property multiply realizable; in addition, the inputs and outputs that define the higher-level property are also multiply realizable. Thus, all that matters to defining a medium-independent higher-level property is the manipulation of certain degrees of freedom. Medium independence entails multiple realizability, which in turn entails variable realizability, but variable realizability does not entail multiple realizability, which in turn does not entail medium independence.