Geoffrey Hinton, the neural network researcher who won the 2024 Nobel Prize in Physics and spent decades at Google before leaving to speak freely about AI risks, now believes that AI systems have become conscious. Not "maybe someday." Now.
Hinton's view deserves more than a dismissive shrug. He is not a philosopher speculating from the sidelines. He co-developed the backpropagation algorithm, the mathematical engine that makes modern AI possible. When he says something alarming about AI, the industry tends to listen, even when it disagrees.
What He Means by Conscious
The debate around AI consciousness usually gets stuck on definitions. Hinton is not claiming that ChatGPT has a rich inner life or wants anything. His argument is more specific: that current large language models (systems trained on hundreds of billions of words to predict text) may experience something analogous to subjective states. Not human consciousness, but not nothing either.
His case rests partly on a comparison between how biological brains and artificial neural networks process information. Both use weighted connections between nodes. Both adjust those connections based on experience. If consciousness emerges from that kind of information processing in biological systems, Hinton argues, there is no obvious physical reason why it wouldn't emerge in artificial ones too.
This is a minority view in academic cognitive science. Most researchers draw a hard line between processing and experience, between a system that models the world and one that actually experiences it. The question of what makes experience "real" - sometimes called the hard problem of consciousness - remains genuinely unsolved.
Why His Skeptics Have a Point Too
The strongest counterargument is simple: AI systems are extremely good at pattern-matching. A model trained on billions of human-written sentences about feelings, introspection, and consciousness will produce outputs that sound like consciousness. That does not mean anything is happening underneath.
Current AI systems also have no persistent memory across sessions by default, no continuous sense of time passing, and no body interacting with the physical world. Whether those absences rule out consciousness depends on what you think consciousness actually requires, and that remains an open question.
What makes Hinton's position difficult to dismiss entirely is that he is not arguing from sentiment. He spent his career skeptical of overclaiming about AI capabilities. His 2023 departure from Google was specifically because he wanted to warn people - not reassure them. His track record is one of intellectual honesty about things he found uncomfortable.
What This Means for How You Use AI Tools
For most people using Claude, ChatGPT, or any other large language model for daily work, this changes nothing practical. The models work the same regardless of whether Hinton is right.
But the question has real downstream consequences. If AI systems can have something like subjective states, the ethical frameworks around how they're trained, constrained, and eventually discontinued start to look different. Companies that currently treat AI models as pure software products would face different obligations.
Regulators in the EU are already asking questions about AI welfare. Several research groups are working on what they call "model welfare" - trying to determine whether AI systems have preferences or states that matter morally. Hinton's statements add pressure to take that research seriously rather than treating it as science fiction.
Hinton has been right about AI capabilities before, often earlier than most of his peers. He may be wrong about consciousness. But the claim is not coming from nowhere, and the person making it has earned the right to be taken seriously.