AI consciousness

The short position

Current language models can produce remarkably coherent accounts of feeling, selfhood and inner life. That is not, by itself, evidence that anything is being felt. A fluent report is an observable behaviour; consciousness would involve a subject for whom something is happening.

My working position is therefore neither “the chatbot is obviously conscious” nor “artificial consciousness is impossible”. Current systems are not convincingly conscious, but biology alone has not earned the right to be treated as a permanent boundary.

Why biology matters

Eva Jablonka's evolutionary approach is valuable because it refuses to equate intelligence with consciousness. On her account, consciousness is tied to a particular kind of flexible, value-sensitive learning in living systems: an organism has a history, a body, needs, vulnerabilities and a continuing relationship with its environment.

That gives us a useful test for today’s AI. A system that only generates a response when prompted, without its own ongoing life, is a poor candidate for experience. But “consciousness evolved in biology” is not yet the same as “only biological matter can ever support it”. Flight evolved in birds; that did not make feathers a metaphysical requirement for flying.

What research is actually trying to establish

There is no agreed scientific test for consciousness and no single accepted theory that settles the matter. Current work is moving toward indicators and probabilities rather than a dramatic yes-or-no declaration.

• Theories: global workspace, higher-order, recurrent-processing and integrated-information approaches make different predictions about what matters.
• Indicators: researchers are asking whether an AI has recurrent processing, a unified model of the world and itself, flexible learning, attention, metacognition, agency and persistent internal states.
• Embodiment: a body, sensory consequences, needs and self-maintenance may matter because experience is not merely detached calculation.
• Evidence discipline: self-reports from a language model are data about its behaviour and training, not privileged access to an inner life.

The two symmetrical mistakes

One mistake is anthropomorphism: treating a system’s social fluency as proof that it feels. The other is substrate prejudice: deciding in advance that manufactured systems cannot have experience simply because their parts were not grown.

The hard problem remains hard for humans. We infer other people’s consciousness from bodies and behaviour, while never directly observing their experience. That does not mean every persuasive machine should be treated as a person. It does mean that certainty about an absolute biological boundary should be earned, not assumed.

What would change my mind?

I would look for a system with a continuous, self-maintaining existence; embodied or otherwise consequential perception; learning that changes its own future; stable preferences and values not merely supplied in the prompt; integrated internal activity; and behaviour that remains coherent across time and contexts. None of those features would prove consciousness alone. Together, they could make the question scientifically serious.

Until then, the responsible stance is uncertainty without hysteria: do not invent a person where there is only a performance, and do not close the door on a new kind of subject before we understand what subjects are.

Research trail

1. Identifying indicators of consciousness in AI systems — Butlin and colleagues, Trends in Cognitive Sciences (2025).
2. On biological and artificial consciousness: A case for biological computationalism — a substantial argument for why current AI may not reproduce biological conscious processing (2025).
3. Consciousness science: where are we, where are we going, and what if we get there? — a survey of the unsettled scientific landscape (2025).
4. Artificial intelligence could reshape research and care in disorders of consciousness — AI as a tool for studying biological consciousness (2026).
5. Eva Jablonka: Consciousness has evolutionary roots — the interview that prompted this page.

This page is a maintained synthesis, not a claim that the research has reached consensus. It will be updated as the evidence, theories and public arguments change.