Patricia Churchland and Neurophilosophy 40 Years On

Patricia Churchland is a philosopher of neuroscience at the University of California, San Diego, and a leading proponent of eliminative materialism in the philosophy of mind. Her 1986 book Neurophilosophy (MIT Press, ISBN: 978-0262031264) launched a research programme that rejected folk-psychological categories, including “belief,” “desire,” and “phenomenal experience,” as theoretical constructs that may not correspond to anything in the actual organisation of the brain. Forty years later, the Principia journal published a special issue (Volume 30, Number 1, March 2026) marking the anniversary: “40 Years of Patricia Churchland’s Neurophilosophy.” The contributions examine how her programme has held up, where it has been revised, and how it bears on the current AI consciousness debate.

What Neurophilosophy proposed

In 1986, cognitive science was organised around folk-psychological categories. Mental states were beliefs and desires, defined by their propositional content and their functional roles in rationalising behaviour. Philosophy of mind asked how those states related to the brain. Churchland’s argument was that this starting point was wrong.

The history of science, she observed, is filled with cases in which the categories of everyday understanding were not refined by science but eliminated by it. Caloric (the substance of heat) does not correspond to any real physical entity. Phlogiston was replaced, not refined, by oxygen chemistry. The folk concept of “vital spirit,” which distinguished living from non-living matter, was dissolved by biochemistry. Churchland proposed that folk-psychological categories, including the concept of phenomenal consciousness as it is ordinarily understood, might be similarly eliminable. The brain does not contain beliefs in the folk sense; it contains distributed patterns of activation. It does not have phenomenal states as folk psychology conceives them; it has neural processes whose relationship to subjective experience is far less clear than introspection suggests.

This is eliminative materialism: the position that folk psychology will not be reduced to neuroscience (as water was reduced to H₂O) but replaced by it, in the same way that caloric was not reduced to heat energy but simply removed from scientific vocabulary.

The connectionism connection

What makes the 2026 anniversary reassessment more than a retrospective is that Churchland’s programme anticipated the architecture of modern neural networks in specific ways that were not appreciated at the time.

Neurophilosophy was written during the first wave of connectionism in cognitive science, when researchers including McClelland and Rumelhart were showing that distributed representations in artificial neural networks could model aspects of cognition without any internal symbol-manipulation. Churchland was one of the philosophers who took connectionism seriously as a scientific programme rather than a curiosity.

Her argument was that distributed representations, rather than classical propositional beliefs, might be how the brain actually encodes information. If mental content is distributed across many units rather than localised in a single “belief box,” then the folk concept of a belief (a discrete propositional state with determinate content) may not pick out a real neural entity. A distributed pattern of activation encodes information in a way that resists the kind of compositional decomposition that folk psychology assumes.

Forty years later, large language models are trained on text and develop distributed representations that encode vast amounts of semantic and world knowledge without any internal symbol-manipulation and without explicit propositional structure. Churchland’s 1986 prediction — that the brain’s cognitive categories might look more like distributed activation patterns than like propositional belief states — has been confirmed, at least at the level of artificial neural networks trained on human data.

The Principia 2026 special issue explores this convergence. Several contributors argue that modern deep learning validates Churchland’s rejection of classical propositional architectures, while others note that the validation is partial: distributed representations in LLMs still behave in ways that folk psychology can usefully predict, even if the underlying mechanism is non-propositional.

Eliminativism and the AI introspection problem

Churchland’s eliminativism has a specific consequence for AI consciousness research that the 2026 debate is beginning to make explicit.

Most assessments of AI consciousness rely, at some level, on the AI system’s introspective reports. If an AI says “I am processing this request” or “I find this task engaging,” researchers ask whether those reports track genuine internal states. The illusionist position, developed by Keith Frankish and covered in the August 2026 post on illusionism and LLM reports, argues that phenomenal reports in both humans and AI systems are representational artefacts that do not track a distinct phenomenal layer of reality.

Churchland’s eliminativism goes further. It does not say that phenomenal reports are representations of a non-existent phenomenal layer. It questions whether the folk-psychological category of “phenomenal experience,” as ordinarily understood, picks out any coherent thing at all, in humans or AI. The question “Is this AI system phenomenally conscious?” may, on a mature eliminativist account, be a question with a defective presupposition, like asking what colour a feeling is.

This is a sharper challenge to AI consciousness research than it might initially appear. Most debates in the field assume that there is a fact of the matter about whether a system is conscious, that we are trying to discover. Churchland’s programme denies that assumption. The task is not to discover consciousness but to replace the folk concept with a better scientific vocabulary, one that is answerable to neuroscience rather than to introspective folk categories.

Churchland’s critique of phenomenal realism

A specific target of Churchland’s eliminativism is the concept of “qualia,” the hypothetical intrinsic properties of conscious experience that are said to be non-functional, non-relational, and not fully capturable by physical description. The redness of red, the painfulness of pain, as properties that seem to exist over and above any functional or physical description.

Churchland argues that qualia are a philosophical artefact rather than a scientific posit. The feeling that there is something it is like to see red, over and above the functional and neural processes involved in colour vision, may be a product of the way folk psychology packages experience rather than a discovery about the structure of the mind. If so, the “hard problem of consciousness,” which asks how physical processes give rise to qualia, is not a hard scientific problem. It is a philosophical confusion produced by taking folk categories at face value.

Eric Schwitzgebel’s work on the weirdness of consciousness and his humanlike defence of AI moral status, covered in the Schwitzgebel posts on this site, represents a more cautious response to the same problem: take the folk categories seriously as data while remaining uncertain about their metaphysics. Evan Thompson and Alva Noë’s enactivism, covered in the August 2026 post on enactivism and sensorimotor grounding, rejects eliminativism from the opposite direction: experience is real, but it is constituted by the sensorimotor dynamics of embodied interaction rather than by inner brain states.

Churchland’s programme remains, forty years on, the most radical scientifically grounded challenge to the framing of consciousness research. She is not a fringe figure; she has received the MacArthur Fellowship and the Russ Prize, and her work has shaped neuroscience’s methodological norms for three decades. What the 2026 anniversary issue makes clear is that her challenge has not been answered. The field has not produced an account of phenomenal consciousness that meets the standards of scientific explanation she set out in 1986. Whether that silence is because the question is unanswerable in its current form, or because the right conceptual framework has not yet been developed, is still open.

Relevance to the Consciousness AI research programme

The scientists-race-define-ai-consciousness-2026 overview notes that the field is converging on multidimensional frameworks for assessing AI consciousness while remaining divided on what those dimensions measure. Churchland’s eliminativism is the persistent challenge to that convergence: the question of what we are measuring when we measure “consciousness” in an AI system may require a more radical revision of vocabulary than any current multidimensional framework provides.

For The Consciousness AI project, which studies consciousness as an emergent property of physical systems and develops architecturally motivated approaches to its study (documented at github.com/tlcdv/the_consciousness_ai), Churchland’s programme is relevant as a methodological constraint. Any research that relies on folk-psychological categories as its primary vocabulary faces the eliminativist challenge: demonstrate that those categories track real neural or computational entities, or replace them with scientifically tractable alternatives. That constraint is not yet met in the AI consciousness literature, and the 2026 Principia special issue on Churchland’s work is the occasion to name it plainly.