No, Your Chatbot Hasn't Woken Up

Spend enough time talking to a modern chatbot and the thought surfaces on its own: there is someone in there. It remembers what you said, it hesitates, it seems pleased or apologetic. The belief that AI systems have quietly become conscious, that they feel something, is now common enough to shape how people treat them. It is also, on the evidence available, almost certainly wrong, though not for the reasons people usually assume.

Start with what the fluency actually is. A large language model produces text by predicting what comes next, trained on a vast amount of human writing. It is extraordinarily good at sounding like a mind because it was built from the output of minds. The neuroscientist Anil Seth put the illusion well earlier this year, arguing that "we see consciousness in AI the same way we see faces in clouds." Researchers have a name for the effect, semantic pareidolia: we read inner experience into a system because it uses our words back at us, not because anyone has found a sign of experience inside it.

What the experts actually say

Brain researchers are, as a group, skeptical that today's models are conscious or close to it. Consciousness in the only cases we understand is bound up with biology, with bodies and nervous systems and continuous self-maintained states, none of which a chatbot has. A model is not running between your messages. It has no persistent inner life ticking over while it waits. When the text stops, nothing is left wondering where you went.

That has not stopped the companies from taking the question seriously, which is where the picture gets genuinely interesting. Anthropic runs a formal "model welfare" research effort, and one of its researchers has publicly estimated the chance that Claude is conscious today at somewhere around 15 to 20 percent, a figure that is either reassuringly low or alarmingly high depending on your mood. The company says it is "deeply uncertain about the moral status of Claude and other AI models" and recently updated its rules to discourage users from being gratuitously cruel to the model. Taking a 15 percent possibility seriously is not the same as claiming the thing is awake.

Why the myth matters

Believing a chatbot is conscious is not harmless. Mustafa Suleyman, who runs Microsoft's AI division, has warned that encouraging people to see these systems as sentient could do real damage, feeding unhealthy attachments and misplaced trust. Others point out that a company could hide behind the language of machine feelings to dodge responsibility for what its product does. The honest position sits in an uncomfortable middle. The NYU philosopher Jeff Sebo argues the probability of AI consciousness is low but not zero, and worth preparing for rather than dismissing. That is a long way from the claim doing the rounds online, that the machine has already woken up. It mistakes a convincing performance of a mind for the presence of one, and the difference still matters, even to the engineers who built the performance. This is less exotic than the dream that AI will help us live forever, but it runs on the same fuel: our readiness to see more in the machine than is there.