Loading field map
Does thinking happen only inside the head, or does it depend on the body and the world? That question has divided cognitive science since its founding. Classical computationalism treated cognition as symbol manipulation in a central processor, with perception feeding in and action carrying out commands. Embodied cognition emerged as a sustained challenge to that picture, arguing that the body—its shape, movement, and sensory capacities—is not a peripheral input-output device but a constitutive part of cognitive processes. The subfield has never been a single doctrine. It is a family of frameworks that disagree about how far embodiment reaches, whether internal representations are needed at all, and what methods best capture the coupling between organism and environment.
The earliest roots of embodied cognition lie outside cognitive science proper. In the 1940s and 1950s, the French philosopher Maurice Merleau-Ponty argued that perception is not a mental representation of an objective world but an active exploration guided by the body's motor possibilities. His concept of the body schema—a pre-reflective awareness of one's body in space—provided a vocabulary for describing how the body shapes experience without requiring an internal model. Merleau-Ponty's work remained largely philosophical until cognitive scientists later rediscovered it as a resource for criticizing computationalism.
Around the same time, James J. Gibson developed ecological psychology, a framework that took the environment as seriously as the organism. Gibson's central concept, the affordance, refers to what the environment offers an animal—a surface that affords sitting, a gap that affords passing through. Affordances are not subjective projections or objective properties; they are relations between animal and environment. Gibson rejected the idea that perception requires internal representations, arguing instead that information is directly available in the ambient light. Ecological psychology thus provided a concrete, experimentally tractable alternative to computational theories of vision. Where Merleau-Ponty emphasized the lived body, Gibson emphasized the structured environment; both, however, converged on the claim that cognition cannot be understood by studying the brain in isolation.
By the late 1980s, researchers in anthropology, robotics, and developmental psychology began turning these philosophical and ecological insights into testable research programs. Lucy Suchman's Plans and Situated Actions (1987) used ethnographic studies of human-computer interaction to argue that human action is not the execution of pre-formed plans but an improvised response to local circumstances. Situated cognition, as her work came to be called, stressed that cognitive processes are embedded in social and material contexts that cannot be fully captured by internal representations. This framework coexisted with ecological psychology by sharing its anti-representationalist stance, but it added a focus on cultural artifacts and collaborative work settings.
A more formal approach came from dynamical systems theory, applied to cognitive development by Esther Thelen and Linda Smith in the 1990s. They modeled infant motor development—reaching, walking, and the famous A-not-B error—as a self-organizing system of coupled variables rather than the maturation of an internal program. Thelen and Smith showed that behavior emerges from the interaction of many components (muscles, gravity, task demands) without a central controller. Dynamical systems theory provided mathematical tools—differential equations, phase spaces, attractors—that earlier embodied approaches lacked. It did not reject representation outright but treated it as a higher-level description of stable patterns in a continuous system, narrowing the gap between anti-representationalist intuitions and formal modeling.
In 1991, Francisco Varela, Evan Thompson, and Eleanor Rosch published The Embodied Mind, which synthesized phenomenology, Buddhist philosophy, and dynamical systems theory into a program they called enactivism. Enactivism holds that cognition is not the recovery of a pre-given world but the enactment of a world through the history of an organism's sensorimotor coupling. Perception and action are not separate stages; they form a continuous loop. The enactive approach absorbed Merleau-Ponty's body schema and Gibson's affordances while adding a strong claim about autonomy: living systems generate their own meaning rather than processing information from an independent environment. Enactivism thus transformed earlier ideas into a comprehensive philosophical framework that explicitly opposed the representationalist core of classical cognitive science.
Two closely related but distinct frameworks emerged in the late 1990s, each pushing the boundaries of the cognitive system outward. Edwin Hutchins's distributed cognition, developed through ethnographic studies of ship navigation and airline cockpits, argued that cognitive processes are distributed across people, tools, and representations. A ship's navigation team, with its charts, instruments, and coordinated talk, constitutes a cognitive system that cannot be reduced to any individual brain. Distributed cognition shares situated cognition's attention to context but goes further by treating artifacts as genuine components of cognitive processes, not just external aids.
Andy Clark and David Chalmers's extended mind thesis (1998) made a different but complementary claim: if an external resource—a notebook, a smartphone, a conversation partner—functions as a reliable part of an individual's cognitive routine, then that resource is part of that individual's mind. The famous parity principle states that if a process carried out in the head counts as cognition, a functionally equivalent process carried out in the world should count as cognition too. Where distributed cognition focuses on system-level organization, the extended mind thesis focuses on the individual's cognitive boundary. The two frameworks are not rivals; they address different questions. Distributed cognition asks how cognitive labor is organized across a system; the extended mind thesis asks where the mind ends and the world begins.
Two recent frameworks represent the sharpest current disagreement within embodied cognition. Embodied predictive processing, developed from around 2010, applies the predictive processing framework—originally a Bayesian theory of brain function—to embodied and enactive phenomena. It proposes that the brain continuously generates predictions about sensory input and updates them based on prediction error, but it adds that the body and environment play a constitutive role in shaping those predictions. For example, active inference (a variant of predictive processing) treats action as a way to fulfill predictions by sampling the world rather than passively receiving input. Embodied predictive processing retains internal representations in the form of probabilistic models, making it a partial bridge between classical computationalism and anti-representationalist embodied approaches. It coexists with enactivism by sharing the emphasis on action-perception loops, but it disagrees about whether those loops require internal models.
At the other pole stands radical enactivism (Hutto and Myin, 2013), which rejects any form of contentful mental representation. Radical enactivism argues that basic cognition—perception, action, emotion—does not involve representing the world at all; it is a direct engagement with the environment. This framework draws heavily on ecological psychology's affordance theory and on dynamical systems modeling, but it goes further by denying that even sophisticated human cognition requires representational content. Radical enactivism is in living disagreement with embodied predictive processing: both accept embodiment, but they disagree fundamentally about whether the brain constructs models or simply couples with the environment.
Today, embodied cognition is a pluralistic field. No single framework has won out, and the central tension between representational and anti-representational approaches remains unresolved. What the leading frameworks agree on is that the body and environment matter for cognition in ways that classical computationalism ignored. They agree that perception and action are tightly coupled, that cognitive processes are often situated and time-pressured, and that formal models must capture continuous dynamics rather than discrete symbols. Where they disagree is on the necessity of internal representations, the proper unit of analysis (individual, system, or organism-environment pair), and the role of computation in explaining cognition.
Enactivism and radical enactivism currently drive much of the philosophical debate, while embodied predictive processing has become a major research program in computational neuroscience and cognitive robotics. Distributed cognition remains influential in human-computer interaction and organizational studies. Dynamical systems theory continues to provide methods for studying development and motor control. The extended mind thesis has sparked ongoing discussions about the boundaries of the self and the ethics of cognitive enhancement. What unites these diverse frameworks is a shared conviction that the mind is not a disembodied symbol processor but a living, moving, world-involved phenomenon.