Home / Philosophy of Mind

The Hard Problem of Consciousness and AI

A glowing mechanism of channels with an empty dark centre

The hard problem of consciousness is the question of why any physical process is accompanied by experience at all. For AI it means this: even if we could explain everything an AI system does, from how it sorts information to what it says about itself, we might still not know whether there is anything it is like to be that system.

That gap is sharper in 2026, because research is now explaining a great deal of what happens inside AI models. Between October 2025 and July 2026 Anthropic researchers reported limited self-monitoring, emotion-like representations and a workspace-like structure in Claude. Each concerns one of the problems Chalmers called "easy", and each team said it had not shown that anything is experienced.

What the hard problem of consciousness is

The philosopher David Chalmers named the problem in 1995, in "Facing Up to the Problem of Consciousness". He wrote that the really hard problem of consciousness is the problem of experience. The Internet Encyclopedia of Philosophy puts it as explaining why any physical state is conscious rather than nonconscious.

Chalmers set it against the easy problems. These are not easy in practice; they are "easy" because they seem open to the ordinary methods of cognitive science. His examples include:

  • telling stimuli apart, sorting them and reacting to them;
  • integrating information;
  • reporting mental states;
  • a system accessing its own internal states;
  • focusing attention and deliberately controlling behaviour.

Each of these is a function, and functions can be explained by describing a mechanism. The hard problem is that, once every function is explained, it still seems to make sense to ask why performing them feels like anything.

An open music box with a glowing tendril of light curling up

The hard problem of consciousness and AI in 2026

AI makes the problem concrete, because recent research reads like a list of easy problems being solved inside models:

On Chalmers's account, solving these in a machine does not tell us whether the hard problem has been solved there too, and the researchers themselves say none of the three shows experience. Whether a machine could have experience depends on which theory you accept. 80,000 Hours notes that functionalism, global workspace theory, higher-order theories and integrated information theory could allow digital minds to be conscious, while biological theories and some forms of dualism could rule them out.

A closer look: the "mapping problem"

In September 2026 a team including Anil Seth, Murray Shanahan and Shane Legg proposed a practical way round the impasse in From cacophony to hierarchy. They separate the hard problem, the metaphysical question of what consciousness is, from a "mapping problem": which features of a system's organisation go with which experiences. They set the first aside and assume, as a working method rather than a claim about reality, that such a mapping exists. That lets researchers compare AI systems against theories and attach probabilities, as described in how to test if AI is conscious. It does not claim to solve the hard problem itself.

The traditional responses still divide the field. The encyclopedia groups them as eliminativism (there is no further fact to explain), strong and weak reductionism (experience is function, or is identical to certain physical states), mysterianism (we may never know), dualism and epiphenomenalism (experience is non-physical), and dual-aspect theory, neutral monism and panpsychism (something more basic underlies matter and experience). Applied to AI they give very different answers. The Chinese room argument makes a related point about meaning.

Checklist: spotting the hard problem in an AI claim

  • Is the claim about a function or about experience? "It reports its internal state" is an easy-problem claim.
  • What theory bridges the gap? Any move from function to experience assumes one of the responses above.
  • Is uncertainty admitted? The 2025 and 2026 studies all stop at function.

More in Philosophy of Mind · Report a correction