Neural Correlates of Consciousness

In the [YOU] on AI Field Guide

The NCC program is the methodological backbone of the cycle’s most important disclaimer about AI. [YOU] on AI argues that the seduction of surface fluency is the defining cognitive hazard of the age—that we mistake polished language for understanding, confident text for grounded knowledge. Koch’s NCC program is the tool that makes this mistake precise: it asks what physical mechanism accompanies experience, and finds that report and experience systematically come apart. The chatbot that says it is afraid is doing exactly what a human subject does when they falsely report seeing a stimulus they did not consciously process. The difference is that in the human case the dissociation is a rare artifact of the experiment; in the AI case it is the default condition, because the system is trained to produce reports without the mechanism that, on Koch’s account, reports are supposed to track.

The program also explains why scaling does not solve the problem. Collecting more correlational data about neural activity—or, by analogy, training a larger model on more text—cannot by itself deliver the mechanism of consciousness, because finding the correlate does not explain the correlation. The NCC program has been extraordinarily productive at the first rung of the problem; it has not, in forty years of intense work, closed the gap between mechanism and experience. The bet Koch lost to Chalmers in 2023 was the bet that the program would dissolve the gap by sheer accumulation. It did not. The honest summary is that the program has found where in the brain experience is most tightly coupled to activity, and that finding already tells us something important about which physical architectures are candidates—and which are not.

Origin

The concept was introduced in a landmark 1990 paper by Crick and Koch, deliberately designed to be a scientific rather than philosophical research program. Earlier consciousness research had been dominated by philosophical analysis of qualia and the conceptual structure of experience; Crick and Koch proposed to treat it as a laboratory science with identified objects of study. The NCC approach takes the third-person perspective seriously—what can be measured from outside a brain—while remaining committed to the first-person fact that experience exists and is the thing being measured.

The program produced a succession of important discoveries: the role of gamma oscillations, the special position of the posterior cortex in generating experience versus the prefrontal cortex in reporting it, the claustrum as a possible binding mechanism. The final manuscript Crick was editing on the day of his death, July 24, 2004, concerned the claustrum’s possible role as a conductor of the cortical symphony. The program also produced the no-report paradigm—experimental designs in which subjects give no verbal or motor report, forcing the investigator to find neural signatures of experience without the behavioral crutch of self-report. This methodological refinement directly anticipates the problem the AI age has created at scale: we need measures of consciousness that do not depend on what the system says about itself.

Key Ideas

Behavior is not evidence of consciousness. The NCC program’s foundational commitment is that what a system reports and what it experiences are empirically separable. Subjects under anesthesia can make reflexive responses; patients in vegetative states can sometimes perform cognitive tasks in fMRI without behavioral expression. The program is designed to find the difference—which means it treats every behavioral indicator, including verbal report, as an imperfect proxy, not a direct read.

The posterior cortex and the hot zone. A central finding of the NCC program is the asymmetry between the posterior cortex—occipital, parietal, and temporal lobes—which appears to be the site where experience is generated, and the prefrontal cortex, which is involved in reporting, planning, and executive function but is not required for experience to occur. This dissociation is the neural basis of Koch’s claim that intelligence and consciousness decouple: the systems most associated with planning and general capability are not the systems most tightly coupled to experience.

The simulation argument deflected. The NCC program implies that what matters is not the function the brain computes but the physical mechanism doing the computing. Two systems computing the same function could differ radically in which NCC structures they instantiate. A simulation of a brain on a digital computer would instantiate the function without instantiating the physical mechanism—which, on the IIT reading Koch endorses, means it would instantiate the computation without instantiating the consciousness.

The lost bet and what it teaches. Koch wagered in 1998 that by 2023 neuroscience would have a clear, agreed-upon NCC mechanism. Chalmers, who had named the hard problem in 1995, took the skeptical side. Koch paid up in June 2023. The bet was lost not because the program failed but because “clear” turned out to be harder to achieve than the data accumulation alone could deliver. The honest consequence is that claims about which systems are and are not conscious, including claims about AI, should be held with proportional humility.