The Hard Problem of Consciousness — AI as a Proxy War

On the eighth of April 2026, Christof Koch took the podium at a BIAL Foundation symposium in Portugal and said, in effect, that the brain might not be generating consciousness at all. Koch had spent decades working on what philosophers call the ‘hard problem of consciousness’. The question of why any physical process is accompanied by subjective experience rather than simply occurring, efficiently, with nobody home. He was not an outsider making this claim. He was Francis Crick’s long-time collaborator, one of neuroscience’s most prominent consciousness researchers, a figure whose career was built on the assumption that if you mapped brain activity precisely enough, you would eventually find the explanation for subjective experience. Now, with the maps more detailed than ever and the explanation still absent, Koch has begun drifting publicly toward a different conclusion: that consciousness might be fundamental to reality rather than manufactured by biology.

What was notable about the moment was not the philosophical position itself – idealists and panpsychists have been making this argument for centuries – but the address from which it was delivered. A figure credentialed by mainstream neuroscience publicly loosened the substrate constraint at an event whose programme included Integrated Information Theory, quantum biology, and the formal questioning of physicalist orthodoxy. The mainstream, in short, is now admitting metaphysical leakage. The question worth asking is not whether Koch is right. It is why that admission matters and why the argument it opens has almost nothing to do with AI.

Or rather, why the argument about AI, which everyone is now having simultaneously, has everything to do with it.

The Hard Problem of Consciousness — Why It Refuses to Dissolve

In 1994, David Chalmers stood at a Tucson conference on consciousness and announced that there were two problems, not one. The easy problems — explaining perception, attention, memory, reportability, the integration of information — were genuinely difficult by ordinary scientific standards, requiring decades of neuroscientific labour. But they shared a common character: they could, in principle, be solved by explaining the relevant functions. Show how the brain processes visual input, discriminates stimuli, generates behaviour, and the easy problems begin to yield.

The hard problem was different in kind. Even a complete functional account of the brain would leave untouched the question that actually disturbs people when they think about consciousness at night. Why is any of this accompanied by experience? Why is there something it is like to see red, rather than the visual processing simply occurring in the dark, efficiently, with nobody home? Chalmers called this the explanatory gap: the space between the third-person account of neural or computational processes and the first-person fact of subjective experience. That gap, he argued, could not be bridged by more neuroscience, because neuroscience explains correlates and mechanisms — it does not explain why mechanisms generate phenomenal states at all.

The internet version of this argument tends to reduce it to “feelings are mysterious.” That is not what Chalmers said. The point is structural. Any explanation of consciousness that works by describing physical or functional organisation leaves a residue — the qualitative character of experience — that the description simply does not touch. You can give a complete account of how the brain represents the wavelength 700nm and drives the linguistic behaviour of saying “red,” and you will still not have answered why that process feels like anything. This is not mysticism. It is a claim about the logical structure of third-person explanation and what it can and cannot reach.

The reason this matters before a single word is said about artificial intelligence is that every position on AI consciousness is already downstream of how one handles this gap. The debate about machines is a debate about consciousness wearing a different costume.

What Must Already Be True

Consider what you need to assume before the claim “a sufficiently sophisticated AI system might be conscious” can even sound plausible.

If you accept some version of computational functionalism — the view that mental states are defined by their functional organisation rather than their physical substrate — then consciousness follows function, and the substrate is irrelevant. Carbon, silicon, or sufficiently organised weather patterns: if the right causal-functional structure is present, consciousness is present. On this view, asking whether an LLM is conscious is a genuine empirical question about functional complexity. Ned Block identified something like this with “access consciousness” — the availability of information for reasoning and verbal report — though he was careful to distinguish it from phenomenal consciousness, the harder target. Daniel Dennett, more aggressively, argued that phenomenal consciousness as Chalmers describes it is itself a kind of illusion, which dissolves the hard problem rather than solving it and allows functional explanation to complete its work unobstructed.

Now consider the alternative. John Searle has spent forty years defending biological naturalism: consciousness is a biological phenomenon, produced by the specific causal powers of biological neural tissue, and it cannot be replicated by running the right program on different hardware. His Chinese Room argument — a person following symbol-manipulation rules produces Chinese output without understanding Chinese — was directed not at AI systems but at the functionalist assumption that syntax is sufficient for semantics, and by extension that computation is sufficient for mind. For Searle, asking whether an LLM is conscious is like asking whether a sufficiently detailed weather simulation is wet. The simulation lacks the causal properties of the thing it models.

Both of these positions — functionalism and biological naturalism — share the assumption that consciousness, whatever it is, must be explained in terms of something else. Either functional organisation or biological causation does the explanatory work. The hard problem is handled, in different ways, by denying its force.

Koch and Giulio Tononi represent a third family of responses. Integrated Information Theory proposes that consciousness is identical with a specific mathematical property — integrated information, expressed as Φ — present to some degree in any system that integrates information in the relevant way. This is a structural theory with panpsychist implications: consciousness becomes a feature of certain physical organisation, not confined to biology, potentially present in very simple systems and absent in others regardless of complexity. A highly modular AI system with low Φ would, on IIT, be less conscious than a thermostat with the right architecture. The framework’s predictions are frequently counterintuitive and its measurement problems are significant, but its philosophical ambition is serious: it is an attempt to give consciousness an objective, substrate-independent grip on the world without reducing it to mere function.

What unites all three positions — functionalism, biological naturalism, IIT — is that each one preloads what will count as evidence before the conversation about AI begins. The functionalist sees a sophisticated language model and asks about causal-functional complexity. The biological naturalist sees a symbol processor and reminds everyone that the room is still Chinese. The IIT theorist asks for the Φ calculation and waits. These are not different answers to the same question. They are different questions, each shaped by a prior decision about where consciousness lives.

The Named Camps and What They Argue About

It is worth being precise about where the live disagreements actually sit, because the public debate conflates several disputes that are philosophically distinct.

The first dispute is metaphysical: is consciousness substrate-dependent, functionally realisable, or fundamental? Searle insists on substrate. Functionalists from Putnam through Dennett deny it. Koch’s current position, alongside panpsychist philosophers like Philip Goff, presses toward fundamentality — consciousness as a basic feature of reality, not derived from anything else. These three positions cannot be adjudicated by neuroscience, because they disagree about what kind of evidence is relevant. More neural correlates do not settle whether consciousness is fundamental any more than more weather data settles whether water is essentially wet.

The second dispute is epistemological: how would we know? The other-minds problem was old before Descartes. We infer consciousness in other humans from behaviour, neural similarity, and evolutionary argument — none of which transfer cleanly to AI systems that share our behavioural repertoire without sharing our biology or evolutionary history. When an LLM produces a sophisticated description of distress, the question of whether anything is distressed cannot be answered by analysing the output. The output is exactly what you would expect whether or not anything is happening inside.

The third dispute, which rarely gets separated from the other two, is ethical: does it matter? If there is some non-zero probability that a sufficiently complex AI system is conscious, does that generate obligations? The answer depends on both the metaphysical question (what kind of thing could be conscious) and the epistemological one (how certain do we need to be). Most public discussion of AI consciousness is actually this third dispute in disguise — concerns about liability, moral status, and what we owe to systems we have built and can turn off — dressed in the language of the first two.

The View From Elsewhere

The substrate-versus-function binary that organises Western debate is not universal, and its parochialism is worth naming before the essay closes.

Buddhist philosophy’s concept of consciousness — particularly in the Yogācāra school — does not begin from a brain-and-body framework in which the question of substrate arises. Awareness is, in certain formulations, the ground condition rather than the product: not something produced by the right kind of matter but the medium in which experience and world appear together. The question “which physical substrate gives rise to consciousness?” does not get the same traction in a framework that does not begin with physical substance as the explanatory primitive. This is not mysticism by another name. It is a different metaphysical starting point — one that, notably, has more structural affinity with idealist and panpsychist positions than with either functionalism or biological naturalism. The fact that Koch’s drift toward questioning physicalism feels transgressive within Western neuroscience should itself register as information about where Western neuroscience started, not about where the argument must end.

What the Debate Is Actually About

The April 2026 Koch moment is best understood not as a scientific breakthrough but as a public record of the pressure the hard problem exerts on anyone who takes it seriously long enough. Koch did not abandon his research programme. He followed its implications. When forty years of searching for the neural correlates of consciousness produces exquisite detail about correlates and persistent silence on why there is experience at all, the options are to declare the question meaningless, to insist the answer is coming, or to reconsider the starting assumptions. Koch has, publicly, chosen the third.

What this means for the AI debate is not that machines are conscious or are not. It means that the argument over machine consciousness is a proxy war conducted on terrain that neither side has surveyed adequately. The functionalist deploys AI’s behavioural sophistication as evidence without settling whether behaviour is the right kind of evidence. The biological naturalist insists on substrate without accounting for why biological neurons escape the explanatory gap that silicon falls into. The panpsychist extends consciousness to the universe but owes an account of why human consciousness has the particular character it has rather than some other. None of these positions is incoherent. Each commits to something before the AI question is asked.

The arguments that appear to be about machines are, beneath the surface, arguments about what consciousness is, where it lives, and whether our explanatory frameworks have any grip on it at all. The AI is not the subject of the debate. It is the occasion for it. And the occasion, like most occasions, flatters itself by believing it is the point.

Frequently Asked Questions

Further Reading and Resources
1. "David J. Chalmers, “Facing Up to the Problem of Consciousness' (Foundational paper)":Still the cleanest statement of the easy/hard problem distinction and the explanatory-gap framework this article inherits.
2. "John R. Searle, “Minds, Brains, and Programs” (Foundational paper): The Chinese Room remains the clearest anti-functionalist challenge in the AI-consciousness debate.
3. "Giulio Tononi & Christof Koch, “Integrated information theory: from consciousness to its physical substrate”(Review article): by Kwame Anthony Appiah: Explores the role of African identity in the global cultural and philosophical landscape.
4. Stanford Encyclopedia of Philosophy: “Yogācāra” (Reference entry): Strongest rigorous bridge for the article’s non-Western section, especially around vijñānavāda and mind-only debates.
5. BIAL Foundation symposium page for Christof Koch’s 8 April 2026 opening lecture (Event / context page): Useful for readers who want the immediate context behind the article’s opening frame; the symposium page explicitly says Koch used extraordinary experiences to question physicalism and defended idealism and panpsychism.