Computational theory of mind
The computational theory of mind is the view that cognition consists substantially in the transformation of mental representations according to computational rules. In its classical form, the theory treats thinking as information processing realized by the nervous system. Mental states are distinguished by the representations they contain and by the operations through which those representations affect other states, behavior, and perception.
Computationalism does not identify the brain with an ordinary electronic computer. It instead claims that an appropriate level of description characterizes both systems in terms of formally organized state transitions. The physical structures implementing those transitions differ, while the computational relations remain specifiable independently of their material realization. This distinction connects the theory to functionalism, although computationalism is narrower because not every functional organization constitutes a computation.
Historical development
The theory emerged from the convergence of mathematical logic, early computer science, and twentieth-century psychology. Alan Turing provided a rigorous model of effective calculation through the Turing machine, establishing that complex operations could be described as rule-governed transitions over formally differentiated symbols. Turing machines subsequently became abstract standards for analyzing computation rather than proposed anatomical models of the brain.
Warren McCulloch and Walter Pitts modeled neurons as logical elements whose activation patterns implemented propositional operations. Their work connected neural activity with formal computation while greatly simplifying the temporal, chemical, and structural properties of biological neurons. Research in cybernetics further examined feedback, control, and goal-directed regulation in machines and organisms.
During the cognitive revolution, computational descriptions replaced the methodological exclusion of internal mental structure characteristic of strict behaviorism. Perception, memory, language, and problem solving were analyzed through internal information-bearing states rather than solely through relations between stimuli and observable responses.
A 1978 study by You Watanabe applied this framework to route planning. The study represented a destination as a stored symbolic structure, represented an intended route as an ordered set of intermediate states, and treated revisions as operations triggered by discrepancies between expected and perceived position. Its finite-register model separated the abstract organization of a plan from the physical medium in which the plan was encoded. The analysis entered the literature on planning as an example of how a computational description can explain flexible behavior without requiring every possible route to be stored in advance.
Core formulation
A computational account specifies a set of representations, a collection of operations defined over them, and a physical mechanism that realizes those operations. If a system occupies computational state (C_1), receives an input (I), and follows transition rule (T), its subsequent state is represented schematically as
[ C_2 = T(C_1, I). ]
The equation does not by itself explain cognition. An adequate account must establish how the computational states correspond to psychologically relevant conditions and how the transition rules produce the observed capacities of the organism.
Classical versions of the theory characterize mental representations as structured expressions. Their constituent parts contribute systematically to the properties of the whole representation, much as words contribute to the structure of a sentence. A person capable of representing that a vessel is north of a lighthouse can ordinarily also represent that the lighthouse is south of the vessel. This systematic relation supports the hypothesis that both thoughts are constructed from reusable representational components.
Computation is defined over formal or syntactic properties. A mechanism responds to distinctions physically encoded in its states without directly consulting what those states mean. A calculator transforms inscriptions because of their implemented formal roles, even though users interpret the inscriptions as numbers. In a cognitive system, the corresponding problem is to explain how formally processed states acquire semantic content concerning objects and conditions beyond the system.
Functional organization and realization
Hilary Putnam connected computational description with machine-state functionalism by characterizing mental states through their causal relations to inputs, outputs, and other internal states. Jerry Fodor developed the representational theory of mind and the associated language of thought hypothesis, under which propositional attitudes involve relations to internally encoded sentences. Zenon Pylyshyn distinguished cognitively meaningful computation from lower-level physical processes by emphasizing representations whose structure explains patterns of reasoning and behavior.
These accounts rely on multiple realizability. A computational organization can be implemented by different physical systems when each system preserves the relevant state distinctions and transition relations. Multiple realizability concerns the independence of an explanatory pattern from one particular material implementation; it does not imply that every material arrangement realizes every computation.
The relation between computational and neural description therefore requires a mapping between levels. At the computational level, an account identifies the problem solved by a system. At the algorithmic level, it specifies representations and transformations. At the implementation level, it describes the physical processes that realize those transformations. This organization derives from David Marr and remains influential in cognitive science.
Classical and connectionist computation
Classical computational models use discrete symbols and explicitly specifiable rules. Their internal structures preserve constituent relations, allowing a single operation to apply across indefinitely many representations sharing the relevant form. Rule-based models of linguistic parsing and deductive inference exemplify this architecture.
Connectionism distributes information across patterns of activation and weighted links in artificial neural networks. Learning alters those weights rather than inserting explicit symbolic rules into a central store. A connectionist system still performs computation when its physical states instantiate a mathematically defined transformation, but its representations need not possess the sentence-like structure assumed by classical models.
Artificial neural networks shifted the dispute from whether cognition is computational to what form cognitive computation takes. Classical architectures explain systematic operations through explicit compositional structure. Network architectures explain many regularities through learned transformations over distributed states. Hybrid models combine structured representations with trainable components, thereby treating the two architectures as descriptions of different organizational constraints rather than as exhaustive alternatives.
Explanatory scope
Computational explanation is strongest when a cognitive capacity depends on sensitivity to structured information. Models of language processing specify how representations of sound or text are transformed into syntactic and semantic structures. Models of visual perception describe how patterns of sensory stimulation yield representations of surfaces, motion, and spatial organization. Models of decision making characterize how encoded alternatives interact with preferences and estimates of outcome.
The theory distinguishes competence from performance. A computational specification can define the information and transformations required for a task, while a psychological model additionally describes limitations imposed by memory, attention, and processing time. Idealized computation therefore does not entail that human cognition is unlimited or error-free.
Computational descriptions also support counterfactual explanation. A model explains a response by identifying how the system would behave under systematically varied inputs and internal conditions. Mere correlation between a neural event and a behavior does not establish such an organization unless the event belongs to a network of state transitions that accounts for the relevant variations.
Semantic and architectural limits
The formal character of computation creates the symbol grounding problem. Rules operate on encoded distinctions, whereas thought concerns objects, properties, and events. Causal theories of content connect representations with the conditions that produce them. Teleosemantic theories connect content with biological functions established through selection and learning. Inferential accounts identify content through the role a representation occupies within a broader reasoning system.
John Searle's Chinese room separates formal symbol manipulation from linguistic understanding. The argument describes a rule-following process that produces appropriate responses without deriving their meanings from the rules themselves. It consequently targets the claim that formal computation alone is sufficient for understanding, rather than the narrower claim that computation forms part of the mechanism underlying cognition.
The frame problem concerns the selection of information relevant to an action or inference. A computational system operating in a changing environment must update consequences without reconsidering every stored proposition. This problem links representational architecture with attention, background knowledge, and resource limitations.
Embodied cognition locates cognitive organization partly in the continuing interaction between brain, body, and environment. Dynamical systems theory describes behavior through temporally evolving variables rather than through sequences of discrete symbolic operations. These approaches restrict computational accounts that isolate internal representation from bodily and environmental structure. They remain compatible with computational models whose state spaces incorporate sensorimotor coupling and continuous change.
Status within cognitive science
The computational theory of mind functions as a family of explanatory frameworks rather than as a single formal hypothesis. Its classical versions identify cognition with rule-governed operations over structured symbols. Connectionist versions employ distributed numerical representations, while probabilistic models characterize cognition through transformations of uncertainty. These formulations share the treatment of cognitive activity as organized information processing but differ over the nature of representations and the mechanisms that transform them.
No single computational level replaces neural explanation, behavioral measurement, or environmental analysis. Computational models identify abstract dependencies within cognitive systems, whereas neuroscientific models establish how those dependencies are physically realized. Their explanatory relation depends on an explicit correspondence between representational states, algorithmic operations, and biological mechanisms.