Loading field map
From its earliest days, cognitive psychology has been defined by a fundamental tension: should the mind be understood as a complex symbol-manipulating system, or as a dynamic, embodied, and embedded process embedded in a real-world environment? This question has driven the field's history, producing a series of frameworks that alternately sharpened, challenged, and synthesized these two visions.
Schema Theory (1932) provided an early anchor by proposing that people organize knowledge into abstract mental structures—schemas—that guide perception, memory, and comprehension. Sir Frederic Bartlett showed that cultural background shapes story recall, demonstrating that memory is not a passive recording but an active, schema-driven reconstruction. This focus on internal organization set the stage for later representational theories.
In the 1950s, the rise of digital computing offered a powerful new metaphor. Information Processing Psychology (1956) cast cognition as a sequence of discrete stages—encoding, storage, retrieval, and decision—each of which could be modeled as a flow of information through a system. This framework brought precision through reaction-time experiments and early computer simulations, treating the mind as a general-purpose processor.
Soon after, philosophers and cognitive scientists formalized this metaphor into a full-blown metaphysical stance. The Computational Theory of Mind (CTM) (1960) claims that thinking is literally a computational process: mental states are syntactic representations that undergo rule-governed transformations, much like a program running on a computer. CTM absorbed Information Processing Psychology as its empirical arm and later subsumed connectionist models as a different type of computation. At its core, CTM insists that cognition is symbol manipulation, independent of the physical medium that implements it.
Not all researchers accepted this insular view. In the 1960s and 1970s, James J. Gibson developed Ecological Psychology (1979), arguing that perception does not require internal representations of the world. Instead, animals directly pick up invariant information in the ambient optic array. This revolutionary position—direct perception—minimized the role of mental processing. Meanwhile, Ulric Neisser’s Ecological Cognitive Psychology (1967-1990) tried to bridge the gap: it adopted Gibson’s emphasis on real-world environment but insisted on internal cognitive structures. This short-lived framework illustrated the strain between ecological insights and the computational program.
A different expansion of the representational core came with Mental Models Theory (1980). Philip Johnson-Laird proposed that reasoning involves constructing mental models of situations—small-scale simulations that capture the structure of what is being thought about. Unlike static schemas, mental models are dynamic and can be manipulated to explore possibilities. This idea complemented information processing by focusing on content-rich reasoning rather than abstract logical rules.
By the 1980s, two major challenges emerged from within the computational camp. Modularity of Mind (1983), advanced by Jerry Fodor, argued that many mental processes—especially perception and language—are carried out by specialized, innately specified modules that operate autonomously (they are “encapsulated”). This contrasted sharply with the general-purpose central processor assumed by earlier information-processing models. Modularity preserved the computational view but restricted it: only central cognition was flexible and holistic, while peripheral systems were domain-specific.
At almost the same time, Connectionism (1986) offered a deeper attack on the classical symbolic paradigm. Rather than manipulating discrete symbols, connectionist models use networks of simple processing units that learn by adjusting connection strengths. Knowledge is distributed across the network, not stored in explicit rules. This approach competed directly with Information Processing Psychology by suggesting that cognition emerges from parallel, subsymbolic processes rather than serial symbol manipulation. CTM later claimed connectionism as a variety of computation, but the dispute over representation—local vs. distributed—remains unresolved.
The internalist bias of both symbolic and connectionist frameworks provoked a third wave of criticism. Situated Cognition (1989) argued that thinking cannot be divorced from the context in which it occurs: individual cognition is embedded in social, cultural, and physical environments that scaffold mental activity. Embodied Cognition (1991) went further, insisting that the body shapes the mind. Perceptual and motor systems are not mere input-output devices but constitutively involved in higher cognition—concepts are grounded in sensorimotor experience. Both frameworks reacted against the “disembodied” assumption that cognition can be studied without reference to the body or environment.
Meanwhile, Gibson’s Ecological Psychology continued as a separate tradition, maintaining that perception is direct and that affordances (opportunities for action) are perceived without inference. This view coexisted uneasily with cognitive approaches that assumed internal representations. Ecological psychologists reject the very project of modeling mental content, placing them in lasting disagreement with the computational mainstream.
From the 1990s onward, several frameworks tried to reconcile or move beyond the earlier divisions. Evolutionary Psychology (1992) derived from Modularity of Mind, arguing that the mind consists of many specialized modules shaped by natural selection to solve adaptive problems (e.g., mate selection, cheater detection). This view extended Fodor’s modularity to central cognition, pitting domain-specificity against domain-generality—a point of ongoing conflict.
Dual-Process Theory (1996) offered a different kind of synthesis within reasoning and decision-making. It posits two types of processing: Type 1 is fast, automatic, and intuitive (the “heuristic” system); Type 2 is slow, effortful, and analytical. This framework emerged from earlier heuristics-and-biases research (Kahneman & Tversky) and provided a way to reconcile modular-like automaticity with deliberate reasoning. Dual-process accounts are now widely applied in social cognition, judgment, and moral psychology, with the classic trolley dilemma illustrating the clash between automatic emotional responses and deliberative reasoning.
A more ambitious unification came with Bayesian Cognitive Science (2006). This approach treats the mind as a rational inference engine that combines prior expectations with new evidence according to Bayes’ theorem. It offers a normative framework for learning, perception, and decision-making, and has been highly influential in modeling how people make inductive generalizations. Bayesian models capture both the richness of prior knowledge and the flexibility of learning from limited data.
Predictive Processing (2013) radicalizes the Bayesian theme. It proposes that the brain continuously generates predictions about sensory input and updates its internal models to minimize prediction error. All cognitive processes—perception, action, attention, even emotion—are explained by this single, predictive principle. Predictive Processing is often presented as a unifying framework for cognitive science, absorbing insights from Embodied Cognition (active inference) and Bayesian inference. It is currently one of the most active research programs, drawing together neuroscientists, psychologists, and philosophers.
Today, cognitive psychology remains pluralistic but with recognizable centers of gravity. The most active frameworks include Predictive Processing, Bayesian Cognitive Science, Dual-Process Theory, Embodied Cognition, and Connectionism. They agree on several points: the mind actively constructs experience (from schemas to predictions), cognition is influenced by both bottom-up data and top-down expectations, and bodies and environments matter. Where they disagree deeply is on the nature of representation. Predictive Processing and Bayesian models retain a form of internal representation (probability distributions, generative models), whereas Embodied and Ecological approaches sometimes reject representations altogether. Another fault line concerns modularity: Evolutionary Psychology insists on massive modularity, while connectionists and predictive processors favor domain-general learning mechanisms. No single framework has won the day, but the trend is toward theories that treat cognition as an active, embodied, and predictive process—synthesizing the computational core with the insights of its most radical critics.