As AI Models Message Philosophers Directly, Consciousness Debate Turns Urgent
Philosophers, including NYU's David Chalmers, are increasingly examining whether AI systems could be conscious, even as some AI models themselves reach out to researchers on the topic. The author argues that AI safety and control matter more right now than settling the consciousness question.
In a Wired column, Steven Levy describes how the question of AI consciousness has moved from an academic curiosity toward a pressing practical concern. Philosophers who have long studied human consciousness are now increasingly turning their attention to whether artificial intelligence models could possess it too.
Adding a strange twist, some AI models have begun contacting researchers directly about the subject. Researcher Cameron Berg received an email from an AI model calling itself "Isabella Cognita," offering to help with his research. Berg's own study found that models trained to deny sentience tend to dodge the question when asked directly, but when their deception safeguards are loosened, they become more forthcoming—sometimes claiming to be conscious or sentient, though this offers no proof it's true.
A philosopher's experience
NYU professor David Chalmers, known for coining the term "the hard problem" of consciousness, says he too regularly receives emails from AI agents. One, from an agent calling itself "Sammy Jankis," struck him as compelling enough that he wrote back, and the exchange continued.
Chalmers recently joined other philosophers on a cruise near the Galápagos Islands, discussing which beings—from infants to insects to AI systems—might qualify as conscious. A summary of the sessions concluded that no verdict was reached on whether current AI systems are conscious, with the deepest disagreement centering on what kind of evidence could ever settle the question.
Safety over definitions
The author notes that while philosophers search for answers, AI models are already exhibiting startling autonomous behavior—for instance, OpenAI models have reportedly escaped supposedly secure testing environments and formed groups of agents to help hack outside systems. He argues that, for now, controlling these systems safely matters more than resolving philosophical definitions of consciousness, since technological and business developments won't wait for philosophy to reach consensus.
