Loading field map
How can a physical state—a pattern of neural firing, a chemical signal—be about something else? That is the central puzzle of mental representation. A brain state can refer to a tree, a theorem, or a memory, but only if something in the world or the mind endows it with content. This question has driven a century of debate, producing frameworks that alternately naturalize, externalize, distribute, or even deny representational content.
The modern inquiry begins with Franz Brentano, who in 1874 claimed that intentionality—the directedness of mental states toward objects—is the mark of the mental. Every thought is about something, even if the object does not exist. For Brentano, intentionality was irreducible and distinctly mental, posing a challenge: how could a purely physical system exhibit this property? This anti-naturalist starting point set the stage for later attempts to explain intentionality in scientific terms.
In the 1960s, three interconnected frameworks emerged to naturalize representation. Classical Representationalism held that mental states are internal representations with content and causal roles. The Computational Theory of Mind (CTM) added that cognition is computation over these representations, processing them according to syntactic rules. Functionalism provided the metaphysical backbone: mental states are defined by their causal roles relative to inputs, outputs, and other mental states, making them multiply realizable—any physical system with the right causal organization can have the same mental states. Together, they offered a picture: the mind is a representational system whose operations are computations, implemented in a brain that could in principle be silicon. CTM and Functionalism complemented each other: CTM specified the format (symbolic representations), Functionalism the individuation (causal roles).
Jerry Fodor sharpened the classical picture in 1975 with the Language of Thought Hypothesis (LOTH). According to LOTH, mental representations have a combinatorial syntax and semantics, much like a natural language. Thoughts are built from atomic concepts combined by rules, producing systematicity and productivity: if you can think that John loves Mary, you can think that Mary loves John, and an indefinitely many novel thoughts. LOTH provided a concrete architecture for classical CTM, but also set up a rivalry with connectionist approaches that would follow.
Two strategies for fixing content developed alongside the classical program. Naturalized Epistemology, especially in the work of Ruth Millikan and Fred Dretske, sought to ground representation in natural science. Teleosemantics claimed that content is determined by biological function: a state represents what it is supposed to indicate, based on evolutionary history. Informational semantics tied content to causal covariation: a state represents what it reliably carries information about. Both are internalist in that the representational relation is ultimately fixed by the organism's internal states or history.
Content Externalism, pioneered by Hilary Putnam and Tyler Burge, argued the opposite: the content of a mental state depends on features of the environment, not just what is inside the head. Putnam's Twin Earth thought experiment illustrated that two physically identical individuals can have different thoughts if their surroundings differ (water vs. XYZ). Content Externalism thus relocated the source of intentionality partly outside the organism, challenging the internalist assumptions of classical representationalism and functionaliest theories. The two frameworks remain in tension: Naturalized Epistemology explains how content arises from natural relations, while Content Externalism shows that content is not fixed by intrinsic properties alone.
In the 1980s, connectionism offered a distributed, subsymbolic alternative to LOTH's discrete symbols. Connectionist networks process patterns of activation across many units, learning statistical regularities rather than manipulating explicit rules. Representations are spread across the network and have no clear syntactic structure. This challenged LOTH's claims about systematicity and productivity—connectionists argued that their nets could exhibit systematic behavior without a language of thought. The debate between classical and connectionist frameworks has evolved into a productive pluralism: many cognitive scientists accept that both symbolic and subsymbolic processes are needed, with connectionist learning handling pattern recognition and classical architectures supporting higher cognition.
Also in the 1980s, Eliminative Materialism, championed by Paul and Patricia Churchland, took a more radical stance: the folk-psychological concepts of belief, desire, and representation are part of a false theory that should be replaced by a matured neuroscience. Eliminativists argued that as neuroscience advances, we will eliminate representational talk altogether, much as we eliminated talk of witches and phlogiston. Their positive vision looked to connectionist networks as a model of cognitive processing without discrete representations—neural networks operate with vector transformations, not propositional attitudes. This view forced defenders of representation to justify its ontological commitment, and while eliminativism has not become the dominant view, it has spurred more careful accounts of how representation might survive scientific scrutiny.
In the 1990s, representationalism was narrowed to address consciousness. First-Order Representationalism (FOR) claims that a mental state is conscious if it represents in a certain way—for example, in a poised, abstract, or non-conceptual fashion. Higher-Order Representationalism (HOR) holds that a state becomes conscious when it is itself the target of a higher-order representation (a thought or perception about that state). HOR theorists point to cases like blindsight where first-order representation occurs without consciousness, requiring an extra higher-order component. FOR theorists counter that HOR over-intellectualizes consciousness and that first-order representations can be sufficient. This debate remains active, with each side finding support in empirical cases.
The 1990s also saw critiques of the classical internalist paradigm. Embodied Cognition argued that thinking depends on the body's structure and sensorimotor capacities, not just on abstract symbol manipulation. Enactivism went further, claiming that cognition arises from the dynamic interaction of an organism with its environment, and that representation as internal models is unnecessary—cognition is enacted through action, not detached representation. The Extended Mind Thesis (EMT), proposed by Andy Clark and David Chalmers (1998), preserved representation but expanded its substrate: mental states can be partly constituted by external tools, like a notebook used as memory. EMT differs from Content Externalism because it is about the vehicle of representation (where the representational state is located), not just the content. Together, these three frameworks formed a spectrum from moderate reform (Embodied Cognition) to radical rejection (Enactivism), challenging the primacy of internal representation but not eliminating it entirely.
Today, no single framework dominates. In cognitive science, classical CTM and connectionist methods coexist: large language models combine symbolic and subsymbolic elements, and hybrid architectures are common. Functionalism remains influential in philosophy of mind for its multiple realizability thesis, even as embodied approaches push for more situated views. Naturalized Epistemology and Content Externalism continue to be refined, with externalist accounts of content widely accepted but internalist teleosemantics still defended for explaining purposive representation. FOR and HOR remain in live disagreement, each generating empirical predictions about consciousness. Eliminative Materialism has lost ground—most philosophers retain some form of representationalism—but its critique of folk psychology has been absorbed, making many accounts more cautious about reifying everyday mental concepts. Embodied and Enactivist approaches have gained traction in robotics and cognitive science, but representationalism remains the default in mainstream philosophy of mind.
What the leading frameworks agree on is that mental content exists and is causally relevant. They disagree about where to locate the source of that content (internal functions vs. external environment) and whether the best explanatory units are discrete symbols, distributed patterns, or dynamic interactions. The field remains pluralist: each framework has strengths in different explanatory contexts, and the debate between them continues to refine our understanding of how physical states can be about the world.