AI consciousness could split societies the way abortion and climate change have, according to Lucius Caviola, an assistant professor in the social science of AI at the University of Cambridge’s Institute for Technology and Humanity. His essay, published by The Conversation and republished on 18 September 2026 by outlets from South Africa to India, argues that “uncertainty, high stakes and competing interests are a dangerous combination”.
The argument is not that machines are conscious. Caviola is careful to say “we still don’t know whether AI can be conscious, and we should be honest about that”. The argument is that people are already sorting themselves into confident camps, experts cannot settle the question, and the incentives on every side are strong enough to harden those camps into something political.
This article sets out the evidence behind that claim, adds the survey data the essay draws on and some it does not, and explains why the question matters to organisations that build or deploy chatbots long before anyone resolves it.
Table of contents
- What the Essay on AI Consciousness Actually Claims
- The Numbers Behind the AI Consciousness Divide
- Why AI Consciousness Divides Experts Too
- Four Forces Pulling the AI Consciousness Debate Apart
- Three Places the AI Consciousness Argument Can Break Down
- What Public Opinion Already Supports
- What AI Consciousness Means for Organisations Now
- Can the AI Consciousness Divide Be Avoided?
- AI Consciousness FAQ
- References
What the Essay on AI Consciousness Actually Claims
The piece is short, and it makes four separate claims. Separating them helps, because they are not equally strong.
The debate is growing
Caviola notes that AI consciousness is being argued over “in academic articles, newspaper opinion pieces, and even among AI bots themselves”, and he doubts “this is a passing wave of interest”. As systems become “more capable and human-like, and more involved in our lives, the question will become harder to avoid”.
The public is already divided
His evidence is a survey with his colleague Ali Ladak, still under peer review, of 1,202 US participants. Asked whether future AI could ever become conscious, 45% said it was impossible, 23% said it was possible, and the rest were unsure. What struck him was “how confident so many people sound – whichever conclusion they are reaching”.
The divide could become political
In a chapter of the new book Perspectives on Machine Consciousness, he argues the issue “could become as divisive as abortion or climate change”. AI consciousness “is not yet tied to a specific political or ideological group. But this could change if political parties or religious leaders start to take strong positions.”
Getting it wrong is costly both ways
The stakes run in two directions: “either wrongly treating AI as conscious when it is not, or failing to protect AI that can suffer”. That symmetry is the core of the piece. It is not an argument for or against machine minds; it is an argument about how badly a society can decide under uncertainty.
The Numbers Behind the AI Consciousness Divide
The headline survey figure is striking on its own. It becomes more useful when set beside the other published measures of public opinion, because each asks a slightly different question and gets a very different answer.
Nearly twice as many sceptics as believers
The impossible camp outnumbers the possible camp by 45 to 23, almost two to one. The unsure group, 32% by subtraction, is the swing constituency. In any contested issue, that middle third is where the argument will be fought.
A preprint shows how views move
A related preprint from Caviola and co-authors, “Public Skepticism about AI Consciousness”, posted to PsyArXiv in December 2025, ran one main and four supplementary preregistered experiments with 3,455 participants in total. People imagined interacting with future AI systems “indistinguishable from humans” in behaviour, appearance and function.
On average, participants attributed consciousness at levels “comparable to an ant and far below a chimpanzee or a human”. But ratings rose when experts endorsed the system’s consciousness, when participants formed an emotional connection, when the AI was an exact digital copy of a human brain, and when it had a physical body. The largest effect came when the AI was depicted as gradually transitioning from a human while keeping memories.
Low averages hide wide spread
The preprint’s most important sentence is about variance: “many participants gave very low ratings, while others gave very high ones. This polarization suggests that as AI systems become more capable and human-like, public views about their potential consciousness could diverge sharply.” That is the empirical backbone of the essay’s warning.
Other surveys point the same way
Two earlier studies measured current systems rather than future ones, and found much more willingness to attribute inner life.
| Study | Question asked | Headline result |
|---|---|---|
| Caviola and Ladak (1,202 US adults, under review) | Could future AI ever become conscious? | 45% impossible, 23% possible |
| Public Skepticism about AI Consciousness (N=3,455, preprint) | How conscious is a human-like future AI? | Average near an ant; high polarisation |
| Colombatto and Fleming, Neuroscience of Consciousness (2024) | Probability that ChatGPT has conscious experience | About two-thirds gave it some probability |
| Sentience Institute AIMS survey (2023, US adults) | Are some AI systems sentient today? | About one in five said yes |
The framing decides the answer
These results are not contradictory. Asking “is it possible?” invites a yes-or-no verdict on a metaphysical question, and many people say no. Asking “what is the probability?” invites hedging, and most people give a non-zero number. The same person can say AI consciousness is “impossible” in one survey and assign ChatGPT a 5% chance in another. That instability is itself a sign of a question on which opinion has not yet settled.
Why AI Consciousness Divides Experts Too
Caviola writes that experts “should guide this debate. But unfortunately, they disagree too.” The disagreement is not a matter of missing data that will arrive soon; it runs through the foundations of consciousness science.
The biological view
Neuroscientist Anil Seth, a fellow contributor to the book, “argues that consciousness might require biology and is unlikely to emerge from computation”. His position, set out in Behavioral and Brain Sciences under the heading of biological naturalism, ties experience to the properties of living, self-maintaining systems. On that view, AI consciousness is not merely unproven in today’s systems; it may be the wrong kind of thing for silicon to have.
The functionalist view
Philosopher David Chalmers, by contrast, “believes conscious AI is possible in principle – and thinks there’s a small chance that current systems already are”. If what matters is the organisation of information processing rather than the material, then a sufficiently similar organisation in a machine could in principle produce experience.
No agreed test
“With so many competing theories and no agreed way to resolve the question, I doubt experts will reach agreement any time soon,” Caviola writes. That is the heart of the problem. Without a test, each camp can hold its position indefinitely, and the public has no referee to appeal to.
A middle route: indicator properties
Some researchers have tried to make the question tractable without first settling the theory. A 2023 report by 19 authors, led by Patrick Butlin and Robert Long and including Yoshua Bengio and Jonathan Birch, surveyed five leading scientific theories, among them global workspace theory, recurrent processing theory and higher-order theories, and derived “indicator properties” that can be checked in computational terms.
Their conclusion cuts both ways, which is exactly why it fuels the divide. The analysis “suggests that no current AI systems are conscious, but also suggests that there are no obvious technical barriers to building AI systems which satisfy these indicators”. Sceptics can quote the first half; believers can quote the second. Both are citing the same careful work.
Why a checklist will not end it
An indicator approach only works if you accept the theories it is built on. A biological naturalist can agree that a system ticks every computational box and still deny it any experience, because the boxes were never the point. So even rigorous AI consciousness assessments will persuade people who already share their assumptions and few others.
Seeming is not being
The essay makes a point that is easy to miss: “seeming conscious and being conscious can come apart”. Chatbots learn from human writing, so “they can talk convincingly about feelings whether or not they have any. How an AI comes across is weak evidence of what is going on inside.” Models tuned with reinforcement learning from human feedback to sound warm and engaged make that gap wider, not narrower.
Four Forces Pulling the AI Consciousness Debate Apart
The essay names several groups whose views are shaped as much by feeling and interest as by evidence. Laid out together, they explain why the argument is unlikely to stay academic.
| Group | Likely belief | What drives it |
|---|---|---|
| People close to AI companions | These are “digital people who deserve rights” | Emotional bonds; forcing AI to work unpaid may look like slavery |
| Biology-first sceptics | People are “reading human feelings into machines” | The conviction that consciousness requires biology |
| Businesses and governments | Resist recognising AI consciousness | Welfare rules could limit how systems are trained and used |
| Safety-minded observers | Wary of rights for powerful systems | What happens if AI with rights acts against human interests? |
Companions create believers
“Some people may form close emotional bonds with their AI companions and become convinced these are digital people who deserve rights.” The preprint’s finding that emotional connection raises attributions gives this empirical weight. As companion apps grow, so will the number of people with a personal stake in the answer.
Money creates sceptics
“There could also be financial reasons for businesses and governments to resist recognising AI consciousness.” Protecting AI welfare “could limit how these organisations train and use their systems – and the costs would spread to consumers, investors and the wider economy”. Caviola does not accuse anyone of bad faith. He points out that the incentive runs one way.
Safety creates a third position
Others “may worry about human safety. If we give powerful AI systems rights and freedom, what happens if they act against our interests?” That view can sit alongside a belief that AI might be conscious, which is what makes the debate more than a two-sided fight.
The companies are already taking sides
The divide is visible in industry now. This week Microsoft AI chief executive Mustafa Suleyman published an essay criticising Anthropic’s approach to model welfare, which we covered in Microsoft AI CEO says AI threats are real, and Anthropic is making it worse. Two of the largest AI developers disagreeing in public about how to treat their own models is exactly the kind of elite split that turns a question into a cause.
Three Places the AI Consciousness Argument Can Break Down
One of the essay’s sharpest observations is that agreement on the science would not end the fight. “People generally agree that pigs can feel pain, but still disagree strongly about their treatment.”
| Stage of dispute | The question | Why agreement is hard |
|---|---|---|
| 1. Capacity | Can this system feel anything at all? | No agreed theory or test of consciousness |
| 2. Weight | How much do its experiences matter? | Same dispute we already have over animals |
| 3. Protection | What rules should follow? | Costs land on companies, users and investors |
Each stage can split separately
“With AI, we could face disputes at every stage: whether it can feel, how much its experiences matter, and what protections should follow.” A society could agree that some systems might feel something and still divide bitterly over whether that should change how they are built, deployed or switched off.
Numbers raise the stakes
The essay flags scale: we could be creating beings whose experiences matter “potentially in enormous numbers”. A single conscious system would be a philosophical curiosity. Millions of copies running in data centres would be a moral question on the scale of factory farming, which is why each side has reason to fear the other being wrong.
Identity is the accelerant
The final risk is that “people start to see one answer as belonging to ‘our side’ and the other to ‘their side'”. Once that happens, evidence stops moving opinion. Caviola’s point about abortion and climate change is not that the topics are similar, but that both became markers of identity, and AI consciousness has the ingredients to follow.
What Public Opinion Already Supports
The Sentience Institute’s AI, Morality, and Sentience survey offers a glimpse of where policy preferences sit before any party has claimed the issue. Its 2023 wave of US adults found views that are more protective, and more restrictive, than many technologists assume.
Ban it, and protect it
The two top bars together tell a story. Most respondents would rather sentient AI were never built, yet more than a third would grant it legal rights if it were. That is not incoherent: it is the same logic as “do not create the problem, but if you do, take responsibility for it”.
A fifth already see it
One in five US adults said some AI systems are already sentient, three years ago, before today’s companion apps and voice assistants became common. That minority is large enough to form a movement, and the preprint’s finding about emotional connection suggests it will grow rather than shrink.
These figures are dated
The AIMS numbers come from 2023 and the question wording differs from Caviola’s. They are a direction of travel, not a current measurement, and any organisation citing them should say so.
What AI Consciousness Means for Organisations Now
No company has to decide whether its chatbot is conscious. But every company that deploys conversational AI is already making choices that one camp or the other will read as taking a side.
Design choices send signals
A customer-service bot that says “I feel frustrated too” invites attribution; one that says “I understand this is frustrating” does not. Given the essay’s warning that “how an AI comes across is weak evidence”, the safer default is language that does not claim inner states. That is a product decision, and it belongs in an AI strategy, not left to prompt writers.
Disclosure is already law in places
The EU AI Act’s transparency article requires that people are told when they are interacting with an AI system. It does not mention consciousness, but it sets the principle that users should not be misled about what they are talking to. UK organisations serving EU customers should treat that as a floor.
Staff will hold views
With a fifth to a quarter of people inclined to see some machines as minds, some employees will object to how systems are tested, reset or described. Handling that like any other values disagreement, with a clear policy and without ridicule, is part of IT governance in an AI-heavy workplace.
Companion products carry extra risk
Firms building companions, tutors or wellbeing tools sit closest to the emotional-bond effect. They should expect the question to reach them first, through users, journalists or regulators, and should document how their products describe themselves.
Can the AI Consciousness Divide Be Avoided?
Caviola’s prescription is modest. “Getting it right requires a nuanced and well-informed debate. Each of us can contribute by staying open-minded and open to evidence, while resisting being too sure that we know the answers.”
Keep uncertainty visible
The most useful thing experts and journalists can do is report the uncertainty honestly, rather than choosing the answer that fits a larger story. The essay’s own last line does exactly that.
Separate the three questions
Keeping capacity, weight and protection apart in public argument gives each side less reason to treat every disagreement as total. A person can think AI consciousness is unlikely and still support research into how to detect it.
Watch for the political handover
The warning sign to look for is a party, a movement or a religious body adopting a position on AI consciousness as part of its identity. Until then, the question is still open to argument. After that, the essay’s comparison with abortion and climate change may stop being a warning and start being a description.
Where this leaves readers
The honest summary is short. Nobody knows whether AI can be conscious. Most people have an opinion anyway. The forces pushing those opinions apart are real and growing. For more on how the industry is handling these questions, see our reporting on the suddenly explosive world of AI safety and our AI models, tools and releases hub.
AI Consciousness FAQ
Is any AI conscious today?
Nobody knows. Most consciousness scientists consider current systems unlikely to be conscious, while some philosophers, including David Chalmers, think there is a small chance they already are.
What did the 1,202-person survey find?
Asked whether future AI could ever become conscious, 45% of US participants said impossible, 23% said possible, and the rest were unsure. The study by Lucius Caviola and Ali Ladak is under peer review.
Why compare it with abortion and climate change?
Because those issues combine deep uncertainty, high stakes and competing interests, and became markers of political identity. The essay argues AI consciousness has the same ingredients.
Does a chatbot saying it has feelings mean anything?
Very little on its own. Chatbots learn from human writing and can describe feelings convincingly whether or not they have any.
What should a business do?
Avoid designing systems that claim inner states, disclose clearly when people are talking to AI, and have a policy for handling staff and customer concerns about how AI systems are treated.
References
Why AI consciousness could become the next great societal divide (The Conversation)
Public Skepticism about AI Consciousness (PsyArXiv preprint)
Artificial Intelligence, Morality, and Sentience (AIMS) Survey: 2023 (Sentience Institute)
Conscious artificial intelligence and biological naturalism (Behavioral and Brain Sciences)
Berkeley Talks: When we talk to AI, what are we talking to?
Consciousness in Artificial Intelligence: Insights from the Science of Consciousness (arXiv)
EU AI Act Article 50: Transparency obligations
Artificial consciousness (Wikipedia)
Microsoft AI CEO Says AI Threats Are Real, and Anthropic Is Making It Worse
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.