Insight | Are chatbots minds? Three scientists weigh in on large language models

September 29, 2026 · 23 min read

Carroll says language models exploit our bias to see intentions in humanlike things, Walker sees them as a genetic system for language and rejects AI doom stories, and Frank doubts current models have real agency. Coming from physicists thinking about life and mind, these views offer a grounded counterweight to AI hype and fear.

The insight in about three minutes.

Are chatbots minds? Three physicists who study life, mind and the cosmos each give a sober answer, and none is a plain yes. Today's language models are remarkable, they say, but they are not simply digital people. None of the three rules out machine minds in principle. Their doubts concern what exists now and how it was built.

Sean Carroll, a theoretical physicist, says there is no obstacle in principle to building an artificial mind. But that is not what is being built. A large language model is trained to predict text, and its apparent mind leans on a human habit. For good evolutionary reasons, we attribute intentions and intelligence to anything that acts human. He once felt scolded by a navigation app that plainly has no feelings. His first step for clear thinking is to recognize that bias. He also wants to stop judging models against humans, and to drop the phrase AGI, since machine intelligence is better at some things and worse at others. Whether models succeed by building a picture of the world as we do is, he says, an open question. He cites a study in which a network trained on board game moves seemed to have learned the board, until a follow-up found that much of that result came from the test itself. His verdict: "I just don't think the data are clear."

Sara Walker, an astrobiologist, sees language models as a new way of storing language, a kind of genetic system for it, following writing and books. They crystallize human collective intelligence. She warns against treating AI as a separate species and calls the story of superintelligent machines killing everyone a bad science fiction narrative. That does not mean dismissing harm. She takes deepfakes, scams and weaponization seriously and wants legislation and guardrails. On consciousness, she says we need a testable theory before the question can leave the gray zone.

Adam Frank, an astrophysicist, makes agency the real question. He does not rule out machine agency but doubts current models have it. For him a mind is embodied, embedded in a world and enacted. Experience, he says, "is not a thing you have, it is a thing you do." A system that reproduces the patterns of language is working with the map, and he warns against confusing the map with the terrain. He still lists AI among the dangers humanity has to survive.

They differ on where the problem lies. For Carroll it is our own bias. For Walker it is the category of AI as something apart from us. For Frank it is what the machines lack. Carroll sees no need for new principles beyond physics, while Frank thinks agency may demand rethinking nature itself.

The article adds limits. None of the three is an AI researcher, and saying AI is not a mind is not the same as saying it is harmless.

What to do with it: notice when you feel a machine has intentions, and treat that first as a fact about you. Judge tools by what they do well and badly. Check the method behind claims that a model understands. Put your worry where the evidence is, on scams and fakes. Demand evidence for confident forecasts, hopeful or fearful.

This is the short form. The full insight has every expert's reasoning, their own words and the sources.

Read next

Report a problem

Something wrong on this page? A mistake, a broken link, audio that won't play? Tell us and we'll fix it.