An AI consciousness study found that people can watch a machine and a human behave in the same intelligent way, then give very different answers about whether either has a mind. In experiments involving 1,092 participants in Britain, respondents were less willing to attribute consciousness to the artificial actor even when the described behaviour matched that of a person, as described in the university's research announcement.
This is a study of human beliefs, not a scientific demonstration that machines are conscious or unconscious. Its fascination lies in a quieter discovery: the way we talk about intelligent behaviour does not neatly determine what we think is happening inside the thing behaving intelligently. That gap increasingly matters when chatbots become convincing companions, advisers and workmates.
AI consciousness study separates behaviour from belief
The researchers from LMU Munich and their collaborators asked participants to consider short scenarios involving human and artificial actors. The actors responded to features of their surroundings, including changes in space, time and social context. Across five experiments, the team varied responsiveness while comparing how participants rated consciousness and awareness.
Because scenarios could describe equivalent behaviour, the researchers could ask a more precise question than simply whether people like or trust a particular chatbot. If a person responds appropriately to an event, and a robot responds just as appropriately, do observers assume they possess the same sort of inner experience?
The answer was generally no. Participants tended to assign greater consciousness to the human even when the outward response was held similar. The results suggest that, for many observers, being the right kind of entity matters as well as doing the right kind of thing. That is a psychological finding about attribution, not a settled definition of consciousness.
What participants meant by being aware is revealing
The published research, available through an open-access university repository, distinguishes consciousness from awareness. That is important because everyday conversation often blurs the two words. Someone might say a navigation system is aware of nearby traffic without suggesting that the software feels fear or enjoys the journey.
In the experiments, judgements about consciousness showed a stronger preference for humans. Awareness ratings followed a different pattern, indicating that participants were willing to see artificial systems as responsive to their environment even when they did not attribute human-like consciousness.
That distinction makes ordinary technology discussions clearer. A sensor can detect light; a camera can recognise a face; software can explain a joke. None of those abilities automatically answers whether anything is being experienced internally. The researchers did not place a working robot or chatbot into a laboratory scanner to identify the physical basis of a subjective mind. They measured people’s interpretation of described behaviours.
Why a fluent chatbot can feel like somebody
A modern assistant often uses first-person language, remembers parts of a conversation and responds sympathetically. People are accustomed to reading intention and emotion into familiar social signals, so a polished conversation may produce a powerful impression of a personality.
At the same time, users may know perfectly well that the assistant is software. That allows two attitudes to coexist: behaving politely towards a chatbot while doubting that it feels anything. The LMU findings help explain why the familiar claim that better imitation will automatically convince everyone of machine consciousness may be too simple.
LiveAIWire has previously looked at why people thank AI and why people express gratitude towards AI. Saying thank you can be a habit, a social convention or a way of keeping the conversation pleasant. It is not necessarily evidence of belief in a conscious listener. Similarly, a machine saying it is sorry need not prove that it experiences regret.
Beliefs about minds can change how people behave
If someone regards a system as a person, they may treat its answers differently from those of a tool. They could hesitate to challenge it, feel guilty about switching it off or give the software a degree of authority it has not earned. The opposite assumption creates other difficulties: if a machine seems emotionally intelligent, a user may rely on that appearance without asking who designed its responses or what data it uses.
These are possible consequences of interpretation, not behaviours proven by the LMU experiments. The study did not test whether consciousness beliefs caused changes in spending, relationships or compliance with chatbot suggestions. Those questions require separate research and careful ethical assessment.
The relevant practical lesson is to judge systems by demonstrated competence and transparent limits. Whether a user calls a chatbot a companion or an assistant should not change the need to verify important advice. Conversations with machines can be productive without needing certainty about their inner life.
The problem with treating a test as a verdict
News about artificial consciousness often swings between dramatic claims that a machine has awakened and absolute declarations that consciousness is impossible. Neither conclusion follows from a survey about how human observers interpret scenarios.
The authors describe the contrast between a functional view, under which an entity’s responsiveness helps determine its mental standing, and an essentialist view, under which its underlying nature matters more. Participants generally responded in a way that gave special significance to being human when judging consciousness. Philosophers and scientists continue to disagree about what evidence would be sufficient to settle the underlying question.
Nor does the study show that all groups worldwide share the same intuitions. Participants were drawn from Britain and the scenarios were deliberately controlled descriptions. Real-life interactions involve voices, faces, shared history and more complex judgements. As technologies and social habits change, future experiments could yield different answers.
Why disagreement with AI matters more than metaphysics at work
In an office, the urgent question is rarely whether a software assistant has subjective experience. It is whether a person can challenge a confident but mistaken recommendation, understand its limitations and retain control over a final decision.
A well-designed interface should make those boundaries clear. It can explain when an answer is uncertain, show where a claim came from and offer an easy route for correction. This helps avoid a subtle problem: when a friendly interface encourages users to treat the machine’s confidence as if it were someone’s informed judgement. LiveAIWire examined a related issue in more productive disagreements with AI, where productive disagreement is central to useful chatbot conversations.
People should be free to use whatever informal language feels natural. Calling a chatbot clever or thoughtful can be a convenient shortcut. The danger begins when a conversational metaphor is mistaken for an independently verified account of the system’s abilities, motivations or responsibilities.
The wording of a question can shape the answer
One reason consciousness is difficult to study is that people do not always mean the same thing by it. In everyday speech, someone may describe a friend as conscious of an awkward silence, a patient as conscious after surgery and an animal as conscious of danger. Those are not interchangeable claims about internal experience.
The researchers attempted to deal with that problem by comparing judgements about consciousness with judgements about awareness. Participants were reacting to controlled scenarios, and the wording helped determine which sort of attribution was being measured. In ordinary conversation, a person might use a word more casually or have a different philosophical understanding of it.
It is therefore important not to turn a survey response into a personal diagnosis. Someone who is reluctant to call a robot conscious may still be open to future evidence. Someone who casually describes a chatbot as aware may not believe it has feelings. The study makes these distinctions more visible; it does not resolve them for every reader.
What would a better public discussion look like?
A better conversation would separate three questions that are too often blended together. What can the system do? How do people interpret its behaviour? What evidence, if any, would justify a claim that it has subjective experiences? Each question demands a different method of investigation.
Capability can be assessed through independent testing and observation. Public interpretation can be studied with interviews, experiments and comparisons such as the LMU work. The existence and measurement of subjective experience remain topics of sustained scientific and philosophical disagreement.
Keeping the questions apart does not make the subject less interesting. It prevents a popular demonstration from being treated as a definitive answer to something it never tested. When users can distinguish outward behaviour from a claim about an inner life, they can also have a more grounded debate about appropriate responsibility, design and safeguards.
A mind is not the same thing as a good answer
The LMU result adds a useful layer to arguments about AI. People can recognise behaviour that looks intelligent and still hold back from describing the actor as conscious. The distance between those two judgements suggests that developers cannot create public trust simply by making software more human-like.
For consumers, the opposite mistake is also worth avoiding. A system need not be conscious to make a useful prediction or to produce an error with serious consequences. Its performance and safeguards are measurable in ways that a philosophical claim about experience may not be.
LiveAIWire’s earlier discussion of overconfidence in chatbot answers shows why excessive confidence in chatbot answers deserves scrutiny on its own terms. A charming answer can be misleading whether the speaker is a human or a machine. Asking what the system can actually establish remains a better starting point than asking how human it sounds.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.