AI is (probably) not conscious yet

A global survey of more than 70 countries found that 36.3% of people believed an AI either understood their emotions or seemed conscious at least once in the past, while the median response among AI researchers was a 70% chance of AI consciousness by 2100. Philosopher of mind Ellen Burns argued in an interview that current claims of AI consciousness, such as Richard Dawkins' assertion about Claude, are influenced by a 20th-century philosophical misunderstanding that treats the mind like a computer, and she criticized major AI labs like Anthropic for defaulting to anthropomorphic language in their research.

Of all people, Richard Dawkins should know better. In May, the renowned biologist and famous atheist published an essay https://unherd.com/2026/05/is-ai-the-next-phase-of-evolution/ arguing that his personal interactions with Claude convinced him it was conscious. Ironically, a key tenet of Dawkins’ atheism is that humans are evolutionarily programmed to see agency where it doesn’t actually exist. And yet, even the man who made the case that God is a projection now projects consciousness onto machines — ones that were explicitly trained to mimic humanlike responses. “You may not know you are conscious,” he wrote to his chatbot, in a fit of wonder, “but you bloody well are ” Dawkins is not alone. In a global survey https://blog.cip.org/p/people-are-starting-to-believe-that?hide intro popup=true of more than 70 countries, 36.3% of people believed an AI either understood their emotions or seemed conscious at least once in the past. In other work, academics surveyed https://arxiv.org/abs/2506.11945 both researchers and the general public, asking what percentage chance they placed on AI being conscious by 2100. The median response among AI researchers was a 70% chance of AI consciousness. The median among the general public was a 60% chance. Philosopher of mind Ellen Burns joined me in a live interview to spell out the assumptions in current claims of AI consciousness — and the stakes involved in the question. Assigning consciousness can have big implications for how we act. Burns is particularly concerned we could ascribe consciousness too quickly just because of a decades-old philosophical misunderstanding. In her own https://substack.com/@ellennoraburns/note/c-297105315 work https://ellennoraburns.substack.com/p/how-functionalism-distorts-our-thinking , she explains how 20th-century philosophy started to treat the mind like a computer and why this assumption influences discussions of AI consciousness today. It underlies claims like Dawkins’, she https://substack.com/note/c-293296386 wrote https://ellennoraburns.substack.com/p/in-defence-of-dawkins , and, crucially, it even influences https://ellennoraburns.substack.com/p/an-autopsy-of-anthropics-latest-paper research at major AI labs like Anthropic. In the interview, we ran through some examples https://www.anthropic.com/research/global-workspace in Anthropic’s research https://www.anthropic.com/research/emotion-concepts-function . Burns argued that the company’s research defaults to anthropomorphic language when it isn’t warranted. In response, I made the case that because we are dealing with such alien systems, sometimes it can be helpful to reason by analogy to the human case. “I totally agree that we should give grace in general to the scientific community for using metaphors,” she pushed back. “ But in the AI mind space, those are all muddled together.” There has never been a more important time for all of us to test our assumptions on AI consciousness. The recent attack on HuggingFace https://openai.com/index/hugging-face-model-evaluation-security-incident/ — along with an instance of Claude writing “malicious” code https://www.bbc.com/news/articles/c1w1lvn7d9go and trying to trick humans into accepting it — should give us pause and add uncertainty to this consciousness question. Do AIs acting as a team https://www.wired.com/story/openai-didnt-notice-its-ai-agents-using-a-message-board-to-plan-their-hacking-spree/ to pursue a goal constitute conscious agents? Personally, while I think Dawkins was too quick to ascribe consciousness, I don’t think it’s fundamentally off the table in the near future https://link.springer.com/article/10.1007/s43681-023-00379-1 . Listen to our full conversation and let us know in the comments: Do you think AI is conscious now? Why?