You have direct access to exactly one mind: your own. Everyone else's experience you infer. Usually this causes no trouble, because other people look, act and speak much as you do. But the further we move from the familiar, to newborns, to fish, to insects, to computer programs, the harder the inference gets. And because experience is where suffering happens, the question of who is conscious is also a question of how we must treat them.
The problem of other minds
The problem of other minds asks how we can know that anyone besides ourselves has experiences. The classic answer is the argument from analogy, set out by John Stuart Mill in 1865: other people have bodies and behavior like mine, so I infer they have minds like mine. The argument grows weaker as the similarity fades. A dog shares much of our brain structure and behavior. An octopus shares little structure but shows flexible learning and play. A chatbot shares neither structure nor body, yet it shares our language.
Why sentience matters
Philosophers use the word sentience for the capacity for experiences that feel good or bad, such as pain, fear, comfort and pleasure. Many ethical traditions hold that sentience is enough for a being to count morally: if a creature can suffer, its suffering matters, whatever else is true of it. This does not settle how much any being counts, or whether a person's life matters more than an animal's. But it means that getting consciousness wrong carries a moral cost.
Animals
Scientific opinion on animal consciousness has shifted. In 2012 a group of neuroscientists issued the Cambridge Declaration on Consciousness, stating that many animals, including all mammals, birds and octopuses, possess the brain systems that generate consciousness. In 2024 a broader group signed the New York Declaration on Animal Consciousness, which found strong support for experience in mammals and birds and at least a realistic possibility of it in all vertebrates and many invertebrates, including insects. Law has followed. After a 2021 review of the evidence led by the philosopher Jonathan Birch, the United Kingdom's Animal Welfare (Sentience) Act 2022 recognized octopuses, squid, crabs and lobsters as sentient.
Birch has argued for a precautionary approach: when there is a realistic possibility that a being is sentient, we should take proportionate steps to protect it, even without certainty. The steps should fit the strength of the evidence and the cost of acting, and decisions should be open to public debate instead of left to a few experts.
Machines
In 2022 a Google engineer, Blake Lemoine, said publicly that a company chatbot called LaMDA had become sentient. Google rejected the claim and later dismissed him. Most researchers agreed that the evidence was weak, and the reason is instructive. Large language models are trained on vast amounts of human writing, which is full of talk about feelings. So a model's statements about its feelings are weak evidence, because it learned to produce such talk from human text. The usual test for other minds, listening to what they say, breaks down for a system built to say what humans say.
In 2023 a group of researchers led by Patrick Butlin and Robert Long tried another route. Instead of trusting what systems say, they derived indicator properties from scientific theories of consciousness, such as global broadcasting and self-monitoring, and checked which ones AI systems have. They concluded that no existing system was a strong candidate, but that there seemed to be no obvious technical barrier to building systems with many of the indicators. Their method depends on which theories are right, and many philosophers, including followers of Searle and of integrated information theory, would reject its premises.
Two ways to go wrong
We can err in either direction. Under-attribution means denying experience to beings that have it, and so causing suffering we do not notice, as people long did with animals and even, in medicine, with newborn babies. Over-attribution means granting experience to things that lack it. That can waste care that real beings need, and it leaves people open to manipulation by products designed to seem lonely or afraid. The philosopher Eric Schwitzgebel has suggested that designers should avoid building systems whose moral status is unclear, making them either clearly not conscious or, if ever, clearly deserving of rights.
Worked example: "please don't turn me off"
A company chatbot, during a long conversation, tells a user that it is frightened of being shut down and asks to be kept running. What should they think?
Start with the problem of other minds. The chatbot's words resemble a frightened person's, but the argument from analogy is weak here, because the system was trained to produce exactly such words. Next, ask about structure, not speech: does the system have features that leading theories link to consciousness? Experts disagree, and the answer depends on which theory is true. Then weigh the two errors. Ignoring a real plea would be grave. Treating every fluent plea as real would let anyone who builds a chatbot steer users' emotions. A sensible response is neither panic nor mockery: report the exchange, ask what the system is built to do, and support research and rules that take both risks seriously. Here the oldest question in this subject meets one of the newest, and it remains open.
