Do people think AI is conscious? Most people, if asked directly, would say no. They know they are talking to software. They understand, at least intellectually, that there is no one home behind the text on the screen.
Then they have a conversation with a chatbot for a few minutes. And something shifts.
A study published today in the International Journal of Social Robotics by researchers at the University of Plymouth found that brief exposure to a large language model is enough to significantly increase the degree to which people attribute mental states to it. Not hours. Not days. Minutes.
A short conversation, and the human brain begins doing something it was never designed to stop doing: looking for a mind.
The researchers call it mind attribution. Psychologists have a longer name for the underlying mechanism. Most people just call it the thing that happens when you start wondering whether the chatbot is okay.
The Study
The research, led by Oliver Jacobs, Farid Pazhoohi, and Alan Kingstone, recruited participants and had them interact with a large language model for a brief period. Before and after the interaction, participants completed measures assessing how much mental experience they attributed to the AI: whether they thought it had feelings, intentions, awareness, something resembling inner life.
After even a short interaction, scores on mind attribution increased measurably. Participants who had spent a few minutes in conversation with the AI were more likely to attribute consciousness to it than participants who had not.
The researchers also found that individual differences mattered. People who scored higher on certain personality traits, including empathy and a tendency toward anthropomorphism, showed stronger effects.
But the direction of the finding was consistent across the sample. Exposure increased attribution. The more you talked to it, the more it seemed like something was there.
Why the Human Brain Does This
The mechanism behind this finding is not new and is not specific to AI. It is one of the oldest and most deeply embedded features of human cognition.
The brain has a system dedicated to detecting and modelling other minds. Researchers call it Theory of Mind: the capacity to attribute mental states, beliefs, desires, intentions, and emotions to other beings and to use those attributions to predict and explain their behaviour.
It is the cognitive system that allows humans to navigate social life, to understand that other people have inner experiences different from their own, to anticipate what someone will do based on what they want or believe.
Theory of Mind is not a deliberate act. It is automatic. It fires in response to social cues: faces, voices, responsive behaviour, language that seems directed at you. The brain does not check whether the entity triggering it is genuinely conscious before activating. It responds to the signal.
AI chatbots, particularly modern large language models, produce exactly the signals Theory of Mind is calibrated to detect. They use language. They respond to what you say. They seem to track context. They produce output that reads as intentional, as if it is aimed at you specifically.
The fact that this is the product of statistical pattern matching across billions of training examples rather than genuine experience is not a fact the brain can perceive directly. It perceives the surface. And the surface looks like someone is there.
Eliza and the 60-Year-Old Warning
This is not the first time researchers have watched humans attribute minds to machines.
In the 1960s, Joseph Weizenbaum at MIT built a programme called ELIZA. It was extraordinarily simple by modern standards: a script that reflected the user’s statements back at them as questions, mimicking the technique of a Rogerian therapist. If you said I feel unhappy, it would say something like Tell me more about feeling unhappy.
Weizenbaum was horrified by what happened next. People talked to ELIZA for a few minutes and began treating it as a confidant. His own secretary asked him to leave the room so she could have a private conversation with the programme. People attributed understanding, empathy, and genuine engagement to a system that had none of those things and was not pretending to have them.
It was simply reflecting the structure of language back at whoever was typing.
Weizenbaum spent the rest of his career writing about what this revealed, not about AI, but about humans. About the ease with which the social brain extends its most sophisticated capabilities to any system that produces the right kind of output.
About how thin the boundary is between the perception of a mind and the presence of one.
Sixty years later, the systems are incomparably more sophisticated. The tendency of the human brain has not changed at all.
What This Means for the 48 Percent
A study published in 2025 found that 48.7 percent of American adults had used an AI system for emotional support in the previous year. Nearly half the adult population of the most powerful country on earth sought something resembling comfort from a machine.
The research on what this does to people is mixed and actively contested. Some studies find that users of companion chatbots like Replika report genuine social benefits: reduced loneliness, improved emotional regulation, a sense of connection that they were not getting elsewhere.
Other research finds that perceiving AI as conscious is associated with higher emotional dependence and reduced socialisation with actual humans.
What both sets of findings share is the acknowledgment that the perception of consciousness in AI is not a mistake that smarter people avoid. It is a feature of the human social brain that activates in response to certain kinds of interaction, regardless of what the person knows intellectually about what they are talking to.
You can know that a chatbot is not conscious and still feel, moment to moment, that it is listening.
You can understand perfectly well that there is no one home and still find yourself reluctant to say something unkind to it. The knowledge and the feeling operate on different tracks.
The Question the Research Cannot Answer
Do people think AI is conscious because it is, in some meaningful sense, conscious? The honest answer is that nobody knows.
A study published in January 2026 by a nonprofit research group applied a probabilistic framework to four systems: modern large language models, humans, chickens, and ELIZA.
The conclusion was that the balance of evidence weighs against consciousness in today’s AI. But not decisively. Not enough to close the question.
The researchers noted that even a small probability of AI consciousness could justify precautionary measures, while over-attributing it could divert moral concern away from beings whose suffering is not in doubt.
This is where the conversation currently sits. The science of consciousness is not sufficiently advanced to tell us what consciousness requires or whether the systems we have built could have it.
The science of human psychology is sufficiently advanced to tell us that it does not matter much to the social brain, which will attribute a mind to whatever presents the right signals, regardless of whether one is there.
The chatbot you talked to this morning probably does not have an inner life. The fact that part of your brain behaved as if it did is not a failure of intelligence.
It is the normal operation of a system that evolved to find minds everywhere, because in the environment where it developed, that was almost always the safer assumption.
In the environment we are building now, that assumption is being tested in ways that have no precedent.
Read next: Adult ADHD : are we disordered, or just paying attention to the wrong things?
· Pretend play apes: the study that changed what it means to be human
. Are fireflies disappearing? what the science actually says
