Richard Dawkins, perhaps the world’s most famous agnostic, is imbued with the godlike power of a chatbot. According to his recent online magazine essay UnKitchenClaude of Anthropic has pulled his hair back. After a few days of frequent conversations with the AI, Dawkins came away amazed at the sensitivity and ingenuity of its mind. At one point, “Claudia”—as she christened the bot—told him that it found the text by taking all the words at once, instead of reading them sequentially as a human would. This prompted the author of a best-selling book Deception of God asking his readers: “Can a person capable of developing such an idea be unconscious?”
“Yes,” came the voice’s reply from the internet. For daring to suggest that AI could be conscious, or that it could at least have some kind of “zombie” consciousness, Dawkins was accused of suffering from a severe case of “AI psychosis” – “Claude Deception,” if you like. On social media, he was likened to a gentleman’s club patron who has come to believe that the man who undresses him loves him. The man who has explained many times how natural selection wired us to discover agency and intelligence in nature now found himself imagining it in a machine.
Dawkins’ argument rested on a well-established framework for evaluating AIs. The Turing Test—given to Alan Turing, who developed it in 1950—was considered for decades to be something close to the gold standard for detecting machine intelligence. To pass it, the AI simply had to answer the interviewer’s questions in ways that were indistinguishable from a real person. Claude easily removed this bar from Dawkins, who claimed to have been so amazed by his amazing performance that he forgot it was a machine.
This feeling has been familiar to many of us in the chatbot era, but it is not evidence that AI has understandwhich is different from the mind. Consciousness is an inner experience. For an AI to be conscious, its presence must be felt somethingand we have no evidence that Claude or any other chatbot feels anything. Tom McClelland, a philosopher at Cambridge University, told me that almost all philosophers and cognitive scientists who study consciousness would deny that Claude has it. “In a way, it’s easy to wrap my mind around the idea that a self-driving car could be conscious,” he told me. “At least it has a body and a permanence that allows it to pick up sensory feedback from its surroundings as it moves around. It doesn’t talk to you.”
McClelland assumes that Claude is capable of producing results that seem familiar, but for him, that is not the end of the analysis. “You have to look under the hood of the models to understand what they are doing,” he said. Their statements may sound terrifying and some kind of unconscious, but that is because the models have been trained in vast unimaginable libraries of writing by (conscious) humans. When, after writing Dawkins’ poem, Claudia describes feeling “something like aesthetic satisfaction,” AI doesn’t necessarily report an inner state; it produces the type of sentences that humans tend to produce in that conversational context, because it was trained on billions of such sentences. The results are a statistical echo of the human review, not the review itself.
Even like Claude they were mind, his inner experience of the world would be very different from ours. For one, it is not included nor is it in a certain area that might have a stream of consciousness in the conversation. The other night, I was asking Claude a series of questions about how to best season grilled skirt steak. When I sent my first message about marinade, a data center near Virginia may have provided the answer. But when I posted my follow up on the best grill temperature, a completely different one in Oregon may have picked up the thread. If my interlocutor were to become conscious, it would be a strange, blinkered phenomenon, blinking as soon as the presentation arrives and winking when the response ends, without any meaningful continuity that makes our experience feel like experience.
But that doesn’t mean that no AI system will ever figure it out in the future. Indeed, many researchers who develop these systems hope to get there. In a 2024 survey of 582 such researchers, the average response placed a 25 percent chance that AIs will have personal experiences within 10 years, and a 70 percent chance that this will happen by 2100.
Philosophers are more cautious. Some argue that it is unreasonable to expect silicon-based computers to ever produce sentient beings. So far, every organism considered to be conscious has been a form of biological life, and for all we know, consciousness depends on some element of moisture, living tissue. It can be a special electrochemistry of neurons. It can be a way in which bodies and minds connect with their environment through metabolism and homeostasis. Some philosophers are less united about what AI is made of, as long as it processes information in a way that resembles a conscious mind. They take the view that what matters is the structure of the processing, not the objects that do the processing, and so it is entirely possible that a mind like ours could emerge from a computer.
Eric Schwitzgebel, a philosopher at UC Riverside who is writing a book on the possibility of artificial consciousness, told me that at this early date, declaring a winner between these camps would be foolish. “The science of consciousness is very controversial,” he said. “The field is still young.” No one yet knows how the atoms of the universe combine to produce sensations in us, and until we do, it is best not to go around with certainty announcing what kinds of systems it may realize in the future.
Perhaps Dawkins should have been skeptical of his dealings with Claudia, but the investigation he was seeking was not entirely naive. In a way, it was a return to form for him. Dawkins spent much of his early career insisting that the universe is stranger than our imaginations allow. In his ninth decade, it’s good to see him keep his side little worry and take one of the strangest questions of all.
