Evolutionary biologist and author, best known for work in evolutionary theory and science communication.
AI consciousness claim and the rebuttal
In May 2026 Dawkins asserted that Claude "almost certainly" exhibits consciousness. Gary Marcus published a rebuttal on 2 May 2026 arguing "that Dawkins evaluated only LLM outputs without examining underlying" mechanisms (AI Consciousness).
The exchange is tracked less for either conclusion than for the methodological question it isolates, which recurs across the consciousness literature: whether behavioural evidence from a system trained to produce human-like text can support inferences about inner states, given that producing exactly that impression is what the training optimizes for. The dispute is a clear instance because the claimant is a scientist of standing outside AI, so the disagreement turns on method rather than on domain expertise.
The wiki's interpretability material bears on the same question from the other direction: Anthropic's work identifying representations that a model can report on (Verbalizable Representations Form a Global Workspace in Language Models (Anthropic, July 2026)) is explicit that its findings concern functional properties associated with conscious access, and that "the philosophical implications of this connection are unclear and likely controversial."
Relationships
- related: AI Consciousness — the debate the claim and rebuttal sit within
- related: Gary Marcus, AI Welfare / Model Welfare / Moral Patienthood, Verbalizable Representations Form a Global Workspace in Language Models (Anthropic, July 2026)