Artificial intelligence can now respond to conversation, images and changing environments in ways that increasingly resemble human behaviour. That resemblance has fuelled a difficult question: when a machine behaves as though it notices what is happening around it, do people also begin to treat it as conscious?
A new peer-reviewed study in Cognition suggests that people draw a surprisingly firm distinction. Across five preregistered experiments involving 1,092 participants in the United Kingdom, researchers found that responsiveness strongly increased judgements of awareness for both humans and artificial agents. Consciousness was different. Even highly responsive artificial agents were judged less conscious than humans showing very little responsiveness.
The distinction matters because words such as conscious, aware, attentive and responsive are often used almost interchangeably in everyday discussion about AI. The new findings indicate that people may actually apply different rules when making these judgements. Behaviour appears to matter greatly for awareness, while consciousness remains tied to beliefs about what kind of entity is being judged.
Testing the same behaviour in humans and machines
The research was conducted by Louis Longin, Jurgis Karpus, Nadine Meertens, Bahador Bahrami and Ophelia Deroy. Rather than simply asking participants whether AI can be conscious, the team designed controlled comparisons in which human and artificial agents could be described behaving in the same way.
Participants read short scenarios involving agents whose responsiveness to their surroundings varied. The situations covered spatial, social and temporal features of the environment. For example, an agent might respond to changes around it or to the emotional state of another person. Crucially, the researchers could keep the described behaviour constant while changing whether the agent was identified as human or artificial.
The project comprised two preregistered sets of studies and five experiments in total. The combined sample was 1,092 UK participants. Depending on the experiment, participants rated consciousness and awareness separately or together. This structure allowed the researchers to test whether the two concepts rose and fell in parallel as an agent became more responsive.
Awareness tracked what an agent did
For awareness, the pattern was largely functional. As responsiveness increased, participants judged both humans and artificial agents as more aware. The human-machine distinction therefore became much less important when participants were evaluating whether an agent noticed or registered its surroundings.
This does not mean that participants treated humans and machines identically in every awareness judgement. At the highest levels of responsiveness, humans were still rated as more aware. The central pattern, however, was that awareness rose steadily with behavioural responsiveness in both kinds of agent.
That result fits what the researchers describe as a functionalist attribution. On this interpretation, people infer awareness primarily from what an entity does. If a system reliably detects relevant features of its environment and responds appropriately, observers become increasingly comfortable describing it as aware.
Consciousness showed a much larger human advantage
Consciousness ratings produced a markedly different pattern. At comparable levels of responsiveness, participants consistently attributed more consciousness to humans than to artificial agents. The separation was strong enough that the most responsive artificial agents were still judged less conscious than the least responsive humans in the experimental comparisons.
The authors describe this as an essentialist pattern of attribution. Behaviour still provides information, but it is not sufficient to erase the distinction between categories of agents. Participants appeared to treat being human as relevant to consciousness over and above the observable behaviour described in the scenario.
This is an important qualification to claims that increasingly human-like AI behaviour will automatically cause people to regard machines as sentient. The experiments show that behavioural similarity does not necessarily translate into equivalent judgements of subjective experience.
Why the distinction between awareness and consciousness matters
The study does not determine whether any AI system actually is conscious. It investigates human judgements about consciousness. That is a separate scientific and philosophical question from establishing whether a machine possesses subjective experience.
What the experiments reveal is that terminology can change how people interpret identical behaviour. A person may be willing to say that an AI system is aware of a change because it detected and responded to that change, while resisting the stronger claim that the system is conscious of it.
This distinction has practical implications for public communication. AI companies, journalists, policymakers and users routinely rely on mental-state vocabulary when describing systems. Phrases suggesting that a model knows, understands, wants or decides something can make complex computational processes easier to explain, but they can also blur the line between functional performance and claims about an inner mental life.
The new evidence suggests that audiences are not simply applying one continuous scale of mind to humans and machines. Instead, the concept being used matters. A comparatively neutral functional term such as awareness can follow observable responsiveness, while consciousness appears to carry additional assumptions about the nature of the entity itself.
A large preregistered sample strengthens the comparison
The study has several methodological strengths. The five experiments were preregistered, reducing the scope for researchers to reshape hypotheses after seeing the data. The total sample of 1,092 participants also provided a substantial basis for comparing attribution patterns across experimental conditions.
Most importantly, the use of matched scenarios addressed a common problem in research on AI perception. If a human and a machine are described doing different things, lower consciousness ratings for the machine could simply reflect weaker or less flexible behaviour. Here, the researchers could compare judgements when the behaviour and circumstances were held constant.
The contrast between consciousness and awareness further strengthens the interpretation. Participants were not uniformly unwilling to attribute mental properties to artificial systems. They readily increased awareness ratings as AI responsiveness increased. Their reluctance was substantially more specific to consciousness.
The findings do not settle machine consciousness
The results should not be interpreted as evidence that humans are correct about whether AI can or cannot be conscious. Public intuitions are not a biological or computational test of subjective experience. The experiments instead describe how people currently reason about artificial and human agents.
The sample also consisted of UK participants, so the strength of the distinction may differ across languages, cultures and populations. Concepts such as consciousness and awareness carry linguistic and philosophical baggage, and translations may not preserve exactly the same boundaries.
Short experimental vignettes are necessarily simpler than prolonged interaction with a sophisticated AI system. A user who develops a relationship with an embodied robot or conversational agent over months may make different judgements from someone evaluating a brief written scenario. Future systems could also behave in ways that exceed the responsiveness levels tested here.
Finally, the study concerns attribution rather than moral status. Whether an entity deserves rights, protection or ethical consideration involves additional questions that cannot be answered solely by asking whether observers describe it as conscious.
A more precise vocabulary for AI
The findings add empirical evidence to a debate often dominated by speculation. People do anthropomorphise technology, but they do not necessarily transfer every human mental property to a machine simply because its behaviour looks human-like.
Across these experiments, awareness behaved largely like a judgement about function: more environmental responsiveness produced more attributed awareness. Consciousness behaved more like a judgement about the underlying kind of entity. Human identity retained a large advantage even when observable responsiveness was matched.
As AI systems become more capable, this distinction may become increasingly important. Describing what a system detects, represents or responds to can communicate its functional abilities without automatically implying that it has subjective experience. The study suggests that many people already make that distinction intuitively, even if everyday language does not always make it explicit.
Source Information
Study: “AI is not as conscious as humans: Essentialism in mental state attributions to artificial systems”
Authors: Louis Longin, Jurgis Karpus, Nadine Meertens, Bahador Bahrami and Ophelia Deroy
Journal: Cognition
Publication date: 1 October 2026
DOI: 10.1016/j.cognition.2026.106733
Study design: Two preregistered sets of studies comprising five experiments, with a combined sample of 1,092 UK participants.
