When AI Behaves Like Us, We Still Do Not Call It Conscious
People can watch an AI system respond to a sound in the room, pick up on someone's mood, and adjust its behavior accordingly. They will say it noticed. They will say it is aware. They will not, as a rule, say it is conscious. A study published this month in the journal Cognition puts numbers to that intuition, and the gap it finds between awareness and consciousness turns out to be stubborn, consistent, and worth paying attention to.
The research comes from a team at Ludwig Maximilian University of Munich, led by Dr. Louis Longin, working with Dr. Bahador Bahrami and Professor Ophelia Deroy. Nearly 1,100 participants took part. The findings, published under the DOI 10.1016/j.cognition.2026.106733, offer what the researchers describe as the first direct comparison of how people attribute the same mental states to AI and humans when both are behaving in exactly the same way.
How the experiment worked
The design was straightforward. Participants read short vignette scenarios, each describing a protagonist responding to its environment. Across conditions, the protagonist was either human or an AI system. Within each condition, the scenarios varied by how responsive the protagonist appeared: some versions showed it reacting clearly to sounds or to the emotional states of people nearby; others showed it largely unresponsive to the same surroundings.
After reading a scenario, participants rated the protagonist on mental state attributes. The ratings covered both consciousness, the sense that the entity has inner subjective experience, and awareness, the sense that it notices and tracks what is happening around it.
Longin describes what makes this design different from prior work: "Whereas previous studies have typically asked general questions about whether AI actually has mental states, our study is the first to directly compare how people attribute the same mental states to AI and humans behaving in exactly the same way, under identical circumstances."
That methodological move matters. It isolates the effect of behavior from the effect of category. When the behavior is held constant, what changes is only whether the actor is human or machine.
Two different stories: awareness and consciousness
The results split cleanly along the awareness-versus-consciousness line.
For both AI and human protagonists, responsiveness drove ratings upward. More responsive protagonists were judged more conscious and more aware than less responsive ones. In that sense, behavior shaped attribution for both categories.
But the similarity ends there when you look at absolute levels.
On consciousness, a clear and consistent gap appeared. Even the most responsive AI protagonist was rated less conscious than the least responsive human protagonist. The behavioral signal moved the needle, but it never closed the categorical difference. For awareness, the picture was different: ratings for AI and human protagonists tracked much more similarly, and in some comparisons the distinction largely disappeared.
Longin summarizes the finding this way: "There is a growing worry in public discussion that people will see AI behaving in a humanlike way and start treating these systems as if they had humanlike mental states. We found something much more nuanced: People are quite willing to say that an AI notices things and is aware of its surroundings. But they clearly draw a line when it comes to the term consciousness. People consistently attribute less consciousness to AI than to humans, even when their behavior is identical."
That nuance is the study's central contribution. The public is not uniformly credulous about AI mental life. It is selective, in a way that tracks which mental concept is on the table.
What this study does not prove
Being precise about what the study shows also requires being precise about what it does not show.
The experiment used vignette scenarios, short written descriptions of behavior, not actual AI systems. Participants were not interacting with a chatbot or watching a robot navigate a room. They were reading. That means the study measures attribution in response to described behavior, which is not the same as attribution in response to experienced behavior.
The ratings are self-reports, collected through a structured instrument. Self-reports capture what participants believe about their judgments, or are willing to say about them, not necessarily the implicit attitudes that might emerge in prolonged interaction.
Most importantly, the study measures human attribution. It tells us about how people think about machine minds, not about whether machine minds exist. A finding that people consistently attribute less consciousness to AI than to humans carries no direct implication for the question of whether AI systems have any form of inner experience. Those are separate questions, and this research addresses only one of them.
Professor Deroy points to a linguistic issue that sits under the attribution data: "We may be using mental terms to refer to AI, but this may only be because we lack better words. After all, AI is trained to act and speak like a human, so these descriptions seem natural."
If that is right, then some portion of the awareness attributions people make may reflect a vocabulary problem rather than a genuine judgment. The words available for describing behavior are human-derived, and applying them to non-human systems may be the path of least resistance rather than a considered mental state assignment.
Why word choice matters for companies, journalists, and policy
The awareness-versus-consciousness split is not just a finding for cognitive scientists. It has direct relevance for how AI systems are described in public, and Bahrami draws that line explicitly: "Our study teaches us lessons for how we communicate about AI. Companies and journalists often reach for hyped descriptions of AI models and credit them with intentions, plans or even moral conscience and hesitation. What our study shows is that everyday judgments are much more discriminating: People do not simply put AI and humans on the same mental scale. But the boundary also depends on which terms we use. For more neutral descriptors such as awareness, the distinction can largely disappear. That makes our choice of language important."
This is a useful corrective on two fronts.
On one side, it challenges the concern that human-like AI behavior will automatically trigger human-like mental state attribution. It will not, at least not for the concept of consciousness. People appear to maintain a categorical resistance there, independent of behavioral cues.
On the other side, it shows that language choice is not neutral. Describing an AI as aware of its environment or attentive to context may elide the very distinction that people would draw if asked directly. A company that consistently reaches for awareness language is working closer to the edge of human attribution than one that uses more technical descriptors.
For policy discussions about AI moral status, rights, or accountability, the specific terminology in play shapes the intuitions people bring to those conversations. Research like this suggests that the field needs to be more deliberate about which mental concepts it uses, and for what purposes.
What to take away
The LMU study does not resolve any of the hard questions about AI consciousness. It does not tell us what AI systems experience, if anything, or where the line between genuine awareness and behavioral mimicry falls.
What it does is document something measurable and replicable: when people evaluate the same behavior performed by a human or an AI, their consciousness attributions diverge significantly, while their awareness attributions stay much closer together. That pattern holds across nearly 1,100 participants and across varying levels of behavioral responsiveness.
The practical implication is narrow but clear. How you describe an AI system's behavior influences how mentally present that system seems to the people listening. "Conscious" and "aware" are not interchangeable in the way public discourse often treats them. Choosing between them is a substantive decision, not a stylistic one.
A system trained to act and speak like a human will invite human-derived descriptions. The question is whether the description chosen reflects what is actually being claimed, or whether it is simply the word that came most easily to hand.
Sources and further reading
TechXplore coverage: https://techxplore.com/news/2026-10-ai-people-conscious-humans.html
Study DOI: https://doi.org/10.1016/j.cognition.2026.106733
Full citation: Louis Longin et al, "AI is not as conscious as humans: Essentialism in mental state attributions to artificial systems," Cognition (2026).
