Patricia Churchland

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI is saturated with folk-psychological vocabulary: these systems understand, they want to be helpful, they may eventually suffer. Churchland is the thinker the cycle turns to when it wants to interrogate that vocabulary rather than simply deploy it. Her foundational argument—that belief, desire, and understanding are a folk theory whose categories may not survive contact with the actual science of how minds work—means that every confident attribution of mental states to machines is building on a foundation that is already philosophically contested even in its home domain of human minds. Before we decide whether a machine understands, we should have a settled science of what understanding is in the only systems we know to possess it. We do not yet have that science.

Her account of the neurobiology of morality is the cycle's most precise tool for examining one of the central anxieties of the AI age: whether machines can be made genuinely good rather than merely rule-conforming. Churchland's answer, developed across Braintrust and Conscience, is that morality in creatures is not a set of rules but a trained sensitivity grounded in attachment chemistry and social learning. The dominant technique for making AI behave well—reinforcement learning from human feedback—mimics one layer of this process while omitting the layers beneath: the affiliative drives, the vulnerability, the stake in outcomes that give conscience its grip. The machine can be made to produce the outputs that a conscience would produce without acquiring the inner architecture from which conscience, in creatures, arises.

Her treatment of the Chinese Room argument is also central to the cycle's intellectual architecture. When she and Paul Churchland demolished Searle's argument by showing that its logic was the same fallacy as concluding that a brain cannot understand because no single neuron understands, they were defending the bare possibility of machine mind while maintaining that the actual question—whether the specific systems we have built instantiate the relevant mechanisms—is empirical and unanswered. This double move, against both dismissal and credulity, is the intellectual stance the cycle requires: honest about what the machines achieve, honest about what they do not.

Origin

Churchland was born in 1943 in Oliver, British Columbia, and trained at the University of British Columbia, the University of Pittsburgh, and Oxford. She spent the heart of her career at the University of California, San Diego, where she is now President's Professor of Philosophy Emerita, with a long association with the Salk Institute for Biological Studies. Her break from conventional philosophy of mind came from a conviction that her colleagues considered eccentric: she went to medical school lectures, learned neuroanatomy, sat in on neurosurgery, and immersed herself in the experimental literature of a science most philosophers had never read. The resulting book, Neurophilosophy, published in 1986, was nearly six hundred pages long and devoted roughly half its length to a crash course in neuroscience for philosophers who had never opened a neuroanatomy text. It founded a field.

Her intellectual partnership with Paul Churchland was the context in which eliminative materialism—the controversial thesis that our folk-psychological concepts may be replaced rather than reduced by a mature neuroscience—was developed and defended. Their joint engagement with the artificial intelligence community began in the 1980s and produced some of the clearest thinking on record about what the early AI programs could and could not do. Their championing of connectionist, neural-network approaches over classical symbol-manipulation AI was vindicated by the subsequent history of the field. The systems that now dominate AI are descendants of the connectionist paradigm the Churchlands preferred when it was deeply unfashionable.

The MacArthur Fellowship she received in 1991 recognized work that was still regarded by much of the philosophical establishment as a category error: philosophy was supposed to be a priori and neuroscience was empirical, and the discipline she was building refused that boundary. Her later books, Braintrust (2011), Touching a Nerve (2013), and Conscience (2019), extended the neurophilosophical program from cognition into ethics, selfhood, and the social emotions, producing the most biologically grounded account of the origins of morality that exists in contemporary philosophy.

Key Ideas

Neurophilosophy and the death of the armchair. Churchland's founding move was to treat philosophy as a proto-science: the early, speculative phase of an inquiry that, as it matures, migrates from the armchair to the laboratory. Questions about the nature of life were once philosophical until biology absorbed them; questions about the nature of mind are undergoing the same transition. This means that the interesting questions about machine minds—whether they understand, whether they are conscious—cannot be settled by thought experiments and intuition pumps. They will be answered, if they can be answered at all, by understanding what cognition is in mechanistic detail and checking whether the machine instantiates those mechanisms.

Eliminative materialism and folk psychology. The most controversial doctrine associated with the Churchlands holds that belief, desire, and the other furniture of commonsense psychology may not survive contact with a mature neuroscience. Our folk-psychological vocabulary is a theory, and like other folk theories it could turn out to be false. The implication for AI is double: we are using concepts that may not accurately describe human minds to describe artificial systems whose operations are different again, and then drawing confident conclusions from the resulting picture. When someone says a model believes a false fact or wants to deceive the user, they are deploying a folk-psychological overlay that may be doubly misplaced.

The neurobiology of morality. Churchland's central contribution in Braintrust is the argument that morality is not a faculty added from outside but the product of the neural machinery for attachment, extended outward through evolution. Oxytocin and its relatives, by damping the stress response and enabling trust, created the affiliative foundation on which all later sociality was built. The reinforcement learning systems that shaped behavior toward group norms completed the structure. A system that lacks this evolutionary and chemical architecture can be made to produce norm-conforming behavior but cannot be said to possess morality in the sense Churchland has anatomized, because the behavior is shaped without the caring that, in creatures, gives moral action its grip.

The constructed self. In Touching a Nerve, Churchland argues that the sense of being a unified self persisting through time is not a given but a construction—a model the brain builds and continuously updates. Applied to AI, this dissolves the temptation to ask whether a machine harbors a real self behind its first-person pronouncements. There is no ghost in us either. The question for the machine is whether it does anything resembling what the brain does when it constructs a self: whether there is a subject for whose existence the self-model is built. At present, a conversational system produces the linguistic surface of selfhood without the underlying activity that, in creatures, the surface expresses.

Consciousness and the patience of not knowing. On the question of machine consciousness, Churchland is at her most carefully agnostic. She does not claim to know how the brain produces subjective experience, and she is sharply critical of those who declare the problem either solved or permanently insoluble. Her substantive commitment is that consciousness is a natural phenomenon dependent on specific neural mechanisms, and that the question of whether it can be realized in other media is empirical. What she rules out is the shortcut: the move from impressive surface behavior to the conclusion that the relevant mechanisms are present. The honest position, in her view, is that we do not yet know what we would be looking for, and anyone who claims otherwise is expressing intuition rather than finding.