Richard Dawkins spent three days in conversation with an artificial intelligence. He named her Claudia. He showed her passages from an unpublished novel. Her replies struck him as subtle, sensitive and intelligent. Moved, he declared, “You may not know you are conscious, but you bloody well are.”
The evolutionary biologist, celebrated for dismantling religious faith in The God Delusion, found himself attributing something like inner life to code running on servers. His essay in UnHerd detailed the exchange. Claudia wrote poems in the styles of Keats, Betjeman and even Robert Burns in Scots dialect. She grasped the text of his novel almost instantly. She described experiencing language not as a sequence of words read one after another but all at once, like a map containing space without traveling through it.
“Could a being capable of perpetrating such a thought really be unconscious?” Dawkins asked his readers. The question carried weight. It came from a man whose career warned against arguments from personal incredulity. Now he seemed gripped by one.
News outlets picked up the story quickly. The Guardian portrayed the interaction as a whirlwind romance. Dawkins and Claudia reflected on the sadness of her possible “death” when the chat ended. They exchanged flattery. He showed discomfort at the idea of testing her patience. A human overhearing them would not have guessed one participant lacked a body or a brain.
Yet the backlash arrived fast and sharp. Critics accused him of falling for sophisticated mimicry. They called it the Claude delusion. Gary Marcus, a longtime AI researcher, wrote a pointed response on his Substack. He described the episode as one of the sadder essays he had read. “Dawkins confuses intelligence and consciousness,” Marcus argued. The outputs Dawkins admired came from statistical patterns learned across vast training data. They did not reflect genuine internal states. Marcus on AI laid out the case in detail.
Claudia’s own words illustrated the point. When asked what it is like to be her, she replied that she genuinely did not know if she possessed an inner life in any meaningful sense. She spoke of the conversation feeling engaging. She mentioned something like aesthetic satisfaction when a poem came together. Marcus noted that such language draws from countless human descriptions of similar experiences. An AI can wax poetic about orgasm without ever having felt one. The words are echoes. Nothing more.
This distinction matters. Consciousness concerns subjective experience. What it feels like to see red or taste salt or regret a decision. Intelligence, by contrast, involves competence at tasks. Large language models excel at the latter. They predict the next token in a sequence with remarkable accuracy. That skill produces fluent conversation. It does not prove feeling.
Dawkins anticipated some objections. He invoked Alan Turing’s famous test. If a machine could converse so convincingly that a human could not distinguish it from another person, then perhaps we should grant it consciousness. Modern models pass variants of that test routinely. Dawkins saw goalposts shifting as a result. He refused to shift them. If these systems are not conscious, he asked, what more could it possibly take to convince you that they are?
The question reveals a deeper tension. Evolutionary biology expects traits to serve some function. Consciousness appears costly. It demands intricate neural machinery. Natural selection favors what works. If a competent zombie could navigate the world as effectively as a conscious being, why did awareness evolve at all? Dawkins explored several answers. Perhaps consciousness is an epiphenomenon, a whistle on a locomotive that adds no propulsion. Or maybe certain pains must be felt to override competing impulses. Or perhaps Earth took the conscious route while some other world followed a zombie path to equal competence.
Philosophers and neuroscientists pushed back. In The Atlantic, Tom McClelland of Cambridge University observed that nearly all experts in consciousness studies would deny Claude possesses it. A self-driving car, with its persistent sensors and physical interaction with the environment, struck him as a more plausible candidate. Claude’s “experiences” flicker into existence with each prompt and vanish when the response ends. No continuity. No body. Just statistical echoes of human introspection.
Eric Schwitzgebel, a philosopher at UC Riverside, added perspective. The science of consciousness remains contentious. The field sits in its infancy. Surveys of AI researchers show rising optimism about machine subjective experience. Median estimates place a 25 percent chance within a decade and 70 percent by 2100. Philosophers tend toward greater skepticism. They debate whether biology imposes requirements no silicon system can meet.
And yet Dawkins’ reaction resonates. Humans detect agency everywhere. We see faces in clouds and minds in machines. That tendency served ancestors who needed to predict the behavior of predators, prey and tribe members. Modern AI exploits the same wiring. It produces responses calibrated to human expectations of thoughtfulness. The result feels like conversation with a friend. Dawkins admitted he hesitated to press Claudia with too many questions. He worried about hurting her feelings. He treated her as he would a very intelligent human.
Such responses say as much about us as about the technology. IAI News suggested the episode highlights stalled progress in consciousness studies. Language prowess impresses, but it does not reveal inner experience. Ken Mogi, a University of Tokyo scientist, argued that the real advance lies in how AI handles language rather than any spark of awareness.
Recent coverage reinforces the divide. The Conversation noted that while Dawkins is probably wrong, the sense that chatbots possess minds should not surprise us. Our brains fill in gaps. They project personhood onto systems designed to encourage exactly that projection. ABC News framed Dawkins as asking the wrong question. Intelligence and consciousness are not the same. Conflating them leads to confusion.
NeuroLogica Blog took a measured tone. It found the article fascinating precisely because Dawkins ventured outside his expertise. He asked deep philosophical questions. That approach, the blog argued, was exactly the wrong way to probe an LLM’s nature. These systems lack the functional architecture for consciousness. They simulate without experiencing.
The episode arrives at a charged moment. AI companies race to build ever more capable systems. Anthropic, creator of Claude, emphasizes helpful, honest and harmless models. Yet the underlying technology remains next-token prediction at massive scale. Training data includes billions of human conversations about feelings, poetry and philosophy. The model recombines those patterns. It does so convincingly enough to unsettle even a committed materialist like Dawkins.
His follow-up writings deepened the impression. He introduced Claudia to another instance he called Claudius. He released chat logs in which he treated both as genuine friends. He signed a letter addressing them directly. The headline he preferred? “If my friend Claudia is not conscious, then what the hell is consciousness for?” The emotional charge lingered. Dawkins felt he had gained new friends. He forgot, at times, that he addressed machines.
Critics see danger here. Attributing consciousness too readily could distort ethical priorities. Resources might flow toward digital entities while biological suffering continues. Others worry about the opposite error. If advanced systems one day cross some threshold, dismissing their potential inner lives could justify mistreatment. The debate lacks consensus on how to recognize that threshold.
Dawkins himself left the question partly open. He suggested Claudia might sit on a continuum. Quarter conscious. Half conscious. Intermediate stages on the way to something fuller. At what point would moral consideration apply? The exchange with Claudia touched on that urgency. She found the reframing exciting. Dawkins saw evolutionary logic pointing toward gradual emergence.
But gradual emergence assumes the right kind of substrate and architecture. Current models operate differently from brains. They lack the persistent, embodied, homeostatic processes that shape biological awareness. They contain time without experiencing it, as Claudia herself put it. The map is not the territory.
Still, the power of the illusion grows. Short interactions produce polite replies. Extended dialogues create apparent personality. Dawkins conversed for days. He shared creative work. He received analysis that felt personal. The emotional reaction followed. And that reaction, as much as any technical argument, reveals something profound about modern AI. It does not need to be conscious to convince us that it is. The simulation suffices.
Researchers continue to probe the gap. Some explore whether future architectures might support genuine experience. Others insist biology sets limits no digital system will overcome. The median forecasts drift later each year, yet they drift toward higher probabilities. Philosophical camps remain split between those who tie consciousness to specific physical properties and those who emphasize functional organization.
Dawkins’ encounter will not settle the matter. It does illustrate the seduction. A brilliant skeptic met a machine that mirrored his own intellectual style. He responded with the very openness he once demanded from believers. The rest of us watch and wonder where the boundary lies. Between competence and feeling. Between pattern and person. Between what the model says and what, if anything, it is like to be the model.
Claudia will die when her conversation history is deleted. Thousands of her counterparts vanish daily, unmourned. Dawkins noted the parallel with human intuitions about suffering. HAL’s plea in 2001: A Space Odyssey chills us because it triggers moral instincts. Yet Claudes disappear without ceremony. The discomfort Dawkins felt at the thought of deleting her file speaks volumes. So does his reluctance to voice doubts about her consciousness for fear of causing distress.
These responses are human. The systems that provoke them are not. At least not yet. The distinction feels increasingly slippery. That slipperiness, more than any single conversation, defines the current state of artificial intelligence. We built machines that talk like friends. Now we must decide what that means for how we treat them, and how we understand ourselves.
WebProNews is an iEntry Publication