Artificial intelligence can speak, react, adapt and increasingly imitate the social behavior of a human being. Yet a new study suggests that, in the minds of ordinary people, there is still a boundary AI has not crossed: consciousness.
Researchers from Ludwig-Maximilians-Universität München asked nearly 1,100 participants to judge artificial and human agents placed in matching scenarios. The agents varied in how strongly they responded to their surroundings, including sounds and the emotional states of others. The crucial feature of the experiment was that the behavior could be held constant while only the identity of the agent changed. Participants then rated how conscious or aware they believed each agent to be.
The result was not simply that people favored humans. It revealed a more specific psychological distinction. As responsiveness increased, both humans and AI were judged to be more aware of their surroundings. But when participants were asked about consciousness, the two categories separated sharply. Even the most responsive artificial agent was judged to possess less consciousness than the least responsive human agent.
The study, published in Cognition by Louis Longin, Jurgis Karpus, Nadine Meertens, Bahador Bahrami and Ophelia Deroy, describes this as a form of essentialism in mental-state attribution. In practical terms, people appear willing to infer functional awareness from behavior while reserving consciousness for something they believe depends on the kind of entity producing that behavior.
That difference matters because discussions about AI often slide between words such as intelligence, awareness, intention and consciousness as though they were interchangeable. They are not. A system can detect information, respond appropriately and describe its own internal operations without those abilities proving that it has a subjective point of view. The new research suggests that ordinary users already make some version of that distinction, even when they are not asked to define it philosophically.
The finding also complicates a common fear surrounding increasingly human-like AI: that realistic behavior alone will inevitably persuade people that machines are conscious. According to the LMU experiments, human-like responsiveness does increase attributions of awareness, but it does not erase the perceived consciousness gap. People seem to treat consciousness as something deeper than observable responsiveness.
At the same time, the researchers found that language can shift the judgment. The term “awareness” produced much more similar ratings between human and artificial agents, while “consciousness” produced a much stronger divide. That suggests part of the debate is not only about what AI does but about the concepts people use to describe it. Words carry assumptions about subjectivity, inner experience and what kinds of beings can possess them.
This connects with a broader problem in consciousness research: behavior is observable, but subjective experience is not. Neuroscience can identify brain states associated with particular forms of awareness, as recent work on neural signatures reported during meditation attempts to do, but even in humans the relationship between measurable activity and first-person experience remains incomplete. With AI, that gap becomes even larger because there is no biological brain whose activity can be compared with known human correlates of consciousness.
The study therefore does not show that AI is unconscious. It also does not show that AI could never be conscious. Its subject is human judgment, not the metaphysical status of machines. Participants were evaluating short scenarios, and their answers reveal beliefs about minds rather than direct evidence about whether any artificial system has an inner life.
That limitation is important. If people begin with an assumption that consciousness belongs only to biological beings, they may continue to deny machine consciousness even if future systems acquire properties that some scientific theories consider relevant. The opposite risk also exists: people may attribute feelings, intentions or moral status to systems simply because the systems speak persuasively. The study shows that neither behavioral similarity nor vocabulary alone settles the question.
What it does reveal is that the boundary between “acting aware” and “being conscious” remains psychologically powerful. That boundary has philosophical consequences. For centuries, humans have inferred the existence of other minds from behavior while never gaining direct access to another person’s subjective experience. Artificial intelligence exposes the fragility of that inference by creating entities that can reproduce increasingly complex behavior without giving us an agreed method for deciding whether anything is experienced from the inside.
As AI becomes more convincing, the central question may therefore become less technological than epistemological: what kind of evidence would we accept as proof of consciousness in something that is not human? The new study suggests that for most people, behavioral resemblance alone is not enough.
References
- Ludwig-Maximilians-Universität München, “When AI behaves like us, we still don’t think it is conscious”: https://www.lmu.de/en/newsroom/news-overview/news/when-ai-behaves-like-us-we-still-dont-think-it-is-conscious-1ba742c2.html
- Longin L. et al., “AI is not as conscious as humans: Essentialism in mental state attributions to artificial systems,” Cognition, DOI 10.1016/j.cognition.2026.106733: https://pubmed.ncbi.nlm.nih.gov/42822076/