Dawkins Talks with Claude: Do LLMs Have Consciousness?

Richard Dawkins, the famous biologist and science popularizer, has published an essay on the possibility of consciousness in language models, written after a two-day conversation with Claude AI. The text appeared on the UnHerd platform and immediately drew the attention of the scientific community. Dawkins himself is known for his skepticism, so his thoughts on AI consciousness sound especially intriguing.

Why Dawkins

An essay on AI consciousness by a world-class scientist has stirred up scientific circles for good reason. Thousands of ordinary people convince each other every day that their AI assistants are all but conscious beings. They post long dialogues with LLMs full of words about soul, reflection, pain and love. But Dawkins's authority gives this discussion weight: when the person talking about AI consciousness is not a Telegram enthusiast but a scientist of this caliber, it becomes a serious conversation.

This is not the first philosophical dispute over the nature of consciousness. But here the subject of the dispute is not people or animals, but an artificial system trained on billions of texts.

The core of the debate

The question of LLM consciousness revolves around several key points:

• Can an LLM have subjective experience if it is a text generator?
• Is complex information processing proof of consciousness, or just a good imitation?
• How do you tell a plausible imitation of reasoning from genuine thinking?
• Can a system without a nervous system be conscious?
• Do we need special criteria for AI consciousness, different from human standards?

Dawkins is probing exactly these boundaries between imitation and reality. He is not alone in this: philosophers split into two camps. Optimists believe AI consciousness is possible; skeptics consider it impossible by definition.

The problem of empty dialogues

The main thing the author of the Habr article criticizes is the gap between form and meaning. People exchange hundreds of messages with an LLM in which words about consciousness and reflection turn up, but the density of actual meaning in these dialogues is close to zero. It is like seeing a pattern in a cloud and taking it as confirmation of the cloud's will.

LLMs really do generate plausible text about complex subjects. But plausibility is not understanding. A model can talk about pain without experiencing it. It can reason about consciousness without having it. Most interesting of all, the model itself does not know it is doing this: it simply picks the next probable word in a sequence. This sets an LLM apart from a human, whose word about pain is usually tied to a real sensation.

What this means

Dawkins's essay matters not for a final answer about AI consciousness, but for the question itself. It is an invitation to the scientific community to work out where the line lies between complex information processing and genuine inner experience. While that line stays blurred, skepticism is the right position. It matters for users too: the better we understand what happens inside an LLM, the lower the risk of attributing to the machine qualities it does not have.