When people compare a human with an artificial-intelligence agent, their judgments about consciousness do not appear to depend on responsiveness alone. A new set of experiments examined how English-speaking participants in the United Kingdom rated matched human and AI agents that differed in how responsive they seemed. The results described in the Cognition article by Longin and colleagues point to a consistent distinction: people treated human agents as more conscious at similar levels of responsiveness, while their judgments of awareness followed responsiveness more similarly across the two categories. The study measures human beliefs about agents; it does not establish whether any AI system is sentient.

What the research asked

Study question: How do people assign mental qualities when the agent in a scenario is presented as human or as AI, and when that agent appears more or less responsive? The researchers compared matched human and AI agents rather than asking participants to evaluate unrelated examples. That matching matters because it gives responsiveness a central role in the comparison while keeping the human-versus-AI distinction visible.

The work focuses on judgments made by observers. Participants read or considered third-person vignettes describing agents and then rated those agents. In other words, the task was about what people believe or infer from descriptions, not about opening a test that could determine the presence of subjective experience. The distinction is especially important for a topic where everyday language can slide between believing that an entity is conscious and demonstrating that it has consciousness.

Observed fact: According to the supplied study facts, participants generally rated humans as more conscious than AI agents when responsiveness was similar. That is a statement about the pattern of ratings in these experiments, not a general claim about every person, every AI system or the nature of consciousness.

How the studies were organized

The research consisted of two preregistered studies containing five experiments. Across the studies and experiments, the sample comprised 1,092 English-speaking UK online participants. The sample size indicates the number of participants reported for this research program; it does not make the sample a universal cross-section of humanity.

Participants encountered third-person vignettes involving matched human and AI agents with differing levels of responsiveness. The comparison therefore had at least two salient features: the kind of agent being described and how responsive that agent appeared. The experiments asked participants to make judgments about the agents, allowing the researchers to compare how those judgments changed as responsiveness changed.

Design fact: The studies were preregistered, which is part of the research description supplied for this article. Scope fact: The participants were English-speaking people recruited online in the UK. These details define the evidence base and should remain attached to any summary of the results.

The use of matched agents also makes the findings easier to state precisely. The study does not simply say that a human example received a higher rating than an AI example. It compares examples arranged around responsiveness, then asks whether the rating gap between human and AI remains. That is why the reported findings include both a category difference and a responsiveness pattern.

What participants judged about consciousness

The clearest reported result concerns ratings of consciousness. At similar responsiveness, participants rated the human agents as more conscious than the AI agents. Put plainly, making an AI agent appear comparably responsive did not make participants treat it as equivalent to a human on this measure.

This finding can be described as a human preference in attribution, or as a category-linked difference in judgment. It should not be expanded into a claim that the experiments measured the true consciousness of the agents. The researchers presented agents through vignettes and recorded participant responses. The data show how the participants judged the described agents under the study conditions.

Responsiveness still mattered. The setup varied responsiveness, so the result is not that participants ignored behavior altogether. Rather, the supplied facts indicate that responsiveness and the human-versus-AI label worked together: agents could be compared at similar responsiveness, yet the human agent still received the higher consciousness judgment.

Careful interpretation: The observation is about ratings, not a ruling on metaphysics. It does not show that all humans are conscious, that all AI systems are unconscious, or that responsiveness is the only feature people use. It shows a pattern in this set of preregistered experiments with this sample and these third-person descriptions.

Why awareness and consciousness were not identical

The reported pattern for awareness was more comparable across human and AI agents. Participants’ awareness judgments tracked responsiveness in a similar way for both categories. As responsiveness changed, awareness ratings changed in a way that was more alike across the human and AI conditions than the consciousness ratings were.

There was still an important qualification: highly responsive humans were rated as more aware. That means the more comparable responsiveness pattern did not remove every difference between the categories. A highly responsive human could receive a higher awareness rating, even while the overall relationship between responsiveness and awareness was described as more similar across human and AI agents.

This distinction helps prevent an overly simple summary. One could say the study found a human advantage in consciousness judgments, but that sentence would leave out the way awareness judgments followed responsiveness. One could also say responsiveness affected both groups similarly for awareness, but that would leave out the higher awareness ratings for highly responsive humans. Both parts belong in a neutral account.

The wording also matters because consciousness and awareness are close concepts in ordinary conversation. In this study, the supplied facts report them as separate judgment measures with different patterns. The results therefore do not justify treating the terms as interchangeable, or assuming that one rating automatically answers the other question.

What the findings do not show

They do not prove or disprove AI sentience

The research measures human beliefs about described agents. It does not directly test whether an AI is sentient, has subjective experience or possesses consciousness in an objective sense. A participant’s rating can be informative about attribution while remaining silent about the inner status of the agent being rated.

They do not establish a universal human response

The participants were English-speaking UK online participants. Their responses provide evidence from that sample, not a universal law governing people in every country, language, age group or setting. The supplied study facts also caution against universal conclusions. Replication with other populations would be needed before extending the pattern beyond the reported sample.

They do not isolate every possible influence

The study compared matched human and AI agents with differing responsiveness, but the facts provided here do not identify every feature of the vignettes or every possible cue participants used. It is therefore safest to describe the reported relationship without claiming that responsiveness alone caused the ratings, or that the human-versus-AI distinction was the only influence.

They do not forecast how future systems will be judged

The experiments report judgments under specified study conditions. They are not a plan, product forecast or estimate of how people will respond to future AI systems. Any discussion of later systems would go beyond the supplied evidence.

Dates, source pages and reporting context

The source information carries more than one date. The Cognition study page is listed with a January 2027 issue date, while the article publication metadata is listed as October 1 and the ScienceDaily item is dated October 13. These are observed publication or page metadata details, not separate findings from the experiments.

The LMU news page is one of the listed sources for the study. The primary article is identified by the DOI for the Cognition publication. Readers should use the article and the linked source pages to check the exact wording, measures and reporting context rather than treating a short summary as a substitute for the full paper.

Keeping the dates and source roles separate is useful. The article, the ScienceDaily report and the LMU page are linked records for the same research item in the supplied materials; the dates do not change the participant sample or the reported pattern of judgments.

Frequently asked questions

1. Did the study find that AI is not conscious?

No. It found that participants rated human agents as more conscious than AI agents at similar responsiveness. Because the task measured human judgments of third-person vignettes, it does not determine whether an AI is or is not sentient.

2. Why did responsiveness matter?

Responsiveness was varied in the matched human and AI agents. The supplied facts report that awareness judgments tracked responsiveness more comparably across the two categories. The findings therefore indicate that behavior-like responsiveness influenced judgments, even though humans retained an advantage in consciousness ratings at similar responsiveness.

3. How large was the sample?

The two preregistered studies and five experiments included 1,092 English-speaking UK online participants. That is the reported sample for this research program. It should not be mistaken for a globally representative population or a guarantee that every observer would respond in the same way.

4. What does preregistered mean here?

Here it is a description of the research design supplied with the study facts: the work involved two preregistered studies. The term identifies how the studies were organized before the reported experiments were completed; it does not turn the results into a measurement of AI consciousness.

5. Are consciousness and awareness the same measure in this report?

No. The supplied facts describe different patterns. Consciousness ratings favored humans at similar responsiveness, while awareness judgments followed responsiveness more comparably across humans and AI, with highly responsive humans still rated more aware. The two measures should therefore be reported separately.

6. Can the findings be applied to every AI system?

Not on the evidence supplied. The study used matched agents presented in third-person vignettes and a sample of English-speaking UK online participants. Its results support a cautious description of judgments under those conditions, not a universal conclusion about all AI systems or all populations.

References

  1. Longin et al., Cognition article, DOI 10.1016/j.cognition.2026.106733.
  2. ScienceDaily, Research on human and AI consciousness judgments.
  3. LMU, When AI behaves like us, we still don’t think it is conscious.

Conclusion

The study presents a measured picture of how people attribute consciousness and awareness. Across two preregistered studies and five experiments, 1,092 English-speaking UK online participants rated humans as more conscious than AI agents at similar responsiveness. Awareness judgments were more comparably tied to responsiveness, although highly responsive humans were still rated more aware. The most defensible conclusion is about human judgment in the reported settings. It is not a verdict on AI sentience, and the sample and vignette design limit broader claims.