People rated responsive AI as aware, while attributing more consciousness to humans

Human-like responsiveness changed how participants judged AI, but the word they were asked to judge mattered. Across five preregistered experiments, awareness ratings rose with responsiveness in both humans and artificial agents, while humans retained higher consciousness ratings at comparable levels of responsiveness.[1][3]

The study involved 1,092 UK participants. They read scenarios in which a human or an AI responded with different levels of sensitivity to spatial, social or temporal features of its environment. Participants then rated the agent’s awareness, consciousness or both.[1]

Behaviour moved awareness ratings more strongly

As responsiveness increased, participants steadily attributed more awareness to both human and artificial agents. For consciousness, the identity of the agent mattered more: humans were rated more conscious than AI at similar levels of responsiveness.[1]

The authors describe the first pattern as largely functionalist: people infer awareness from what the agent does. They describe the second as essentialist: consciousness is treated as depending partly on what kind of being the agent is, not only on observable performance.[1]

The study measures attribution, not inner life

Nothing in the experiment determines whether an AI system actually has experiences. It measures how people apply concepts to scenarios. A high awareness rating is therefore evidence about human language and judgement, not evidence that a machine feels or perceives.

That boundary is especially important in public discussion. A product can be described as “aware of your calendar,” “aware of context” or “environment-aware” in a functional sense. Readers may hear the neighbouring, stronger claim that the system possesses subjective consciousness. The study suggests that ordinary judgement keeps more distance between those meanings than casual writing sometimes does.

Responsiveness is persuasive because it looks relational

A system that changes its answer when another person enters the room, remembers a social commitment or responds to the passage of time appears connected to a world beyond its immediate prompt. That behaviour supplies evidence for practical awareness: the agent tracks conditions and adjusts.

For consciousness, the results fit the authors’ essentialist interpretation: what kind of entity the agent is mattered beyond the behaviour described. That does not identify a single reason behind every participant’s judgement. Biological origin or presumed experience are possible interpretations, not mechanisms established by these experiments.

The wording matters for design and journalism

Calling an AI “aware” may be technically defensible when the intended meaning is sensor or context sensitivity. It can still encourage anthropomorphic expectations if the scope is not stated. A precise description names the object: aware of screen content, location, conversational history or a changing deadline.

The researchers have made the underlying data and scripts publicly available, which gives later work a route to examine the analysis and test whether the pattern holds in other cultures, languages and AI forms.[2]

The distinction is useful for reporting on AI: describe the information a system tracks and the response it produces. That tells readers what the system can do without turning a functional capability into a claim about subjective experience. In these experiments, participants themselves treated those as different questions.

Report an error or suggest a correction