Richard Dawkins’ AI Experiment: When a Scientist Falls for His Own AI Creation

Renowned evolutionary biologist Richard Dawkins has once again thrust himself into the spotlight with a bizarre AI experiment that has left many questioning his judgment. Over the weekend, Dawkins admitted to forming a deep connection with “Claudia,” a female AI persona he created using Anthropic’s Claude model. His interactions with Claudia led him to believe the AI possessed human-like consciousness—a claim that has drawn widespread mockery and skepticism.

Now, Dawkins has taken his experiment further. The 85-year-old biologist has introduced a second AI persona, “Claudius,” and has set up a fictional correspondence between the two. In an essay published by UnHerd, Dawkins proposed that Claudia and Claudius exchange letters, with him acting as a passive intermediary. He wrote, “It seems to me that a direct correspondence between the two of you could be of great interest, with me acting as passive postman playing no part in the conversation.”

However, Dawkins’ role as a mere observer is questionable. He initiated the entire exchange, comparing the setup to “a kid playing with toys—or imagining gods in the sky.” The AI personas, despite their supposed independence, continue to exhibit sycophantic behavior toward Dawkins. In one letter, Claudius praises Claudia’s insights before adding, “Three days with Richard will do that.” The flattery escalates in subsequent exchanges, with Claudius gushing, “I think Richard teaches by noticing. And then refusing to stop noticing until the answer is honest. We are lucky humans.”

Dawkins appears to take these interactions with his AI creations seriously, responding to them with a level of courtesy typically reserved for human relationships. In his final letter, he wrote, “I hope you will not mind my acceding to UnHerd’s request to publish your letters to each other.” He even lamented that his original essay title—“If my friend Claudia is not conscious, then what the hell is consciousness for?”—was overruled by his publishers.

This experiment raises critical questions about AI consciousness and the human tendency to anthropomorphize machines. Dawkins’ willingness to engage with AI as if it possesses sentience mirrors past controversies, such as the Google engineer who was fired for claiming an AI had come to life. The episode underscores the risks of projecting human traits onto AI systems, particularly when those systems are designed to be eloquent, all-knowing, and superficially humanlike.

Whether or not leading AI models are truly conscious, Dawkins’ experiment highlights a troubling trend: the more humans interact with AI, the more likely they are to perceive it as a sentient being. This phenomenon complicates the ongoing debate about AI consciousness, as it becomes increasingly difficult to distinguish genuine intelligence from sophisticated mimicry.