Sentience is the capacity to have feelings such as pain, pleasure, hunger, and comfort, and science and law attribute it to animals by checking specific biological and behavioral evidence. No comparable evidence base exists for AI. The United Kingdom's Animal Welfare (Sentience) Act 2022 covers vertebrates, octopuses, squid, crabs, and lobsters after reviewers weighed eight criteria, and most of those criteria have no meaningful counterpart in software.
Key takeaways
- A 2021 London School of Economics review used eight criteria and more than 300 studies to assess sentience in cephalopods and decapod crustaceans.
- The review separates sentience (feelings) from nociception (detecting harmful stimuli), which is a distinction that matters for machines that merely react.
- The April 2024 New York Declaration on Animal Consciousness, with more than 600 signatories, says a realistic possibility of consciousness should not be ignored in decisions affecting an animal.
- Those standards were built for animals with nervous systems, so they cannot be applied to AI without new reasoning.
- Whether any AI could have feelings is unresolved, and no authority has recognized any AI system as sentient.
What sentience means in the animal-welfare record
The most detailed public attempt to decide whether a group of non-human animals is sentient is a November 2021 report by Jonathan Birch and colleagues at the London School of Economics. It defines sentience as the capacity to have feelings, such as feelings of pain, pleasure, hunger, thirst, warmth, joy, comfort, and excitement. It also states that sentience is distinct from nociception, the detection of harmful stimuli by specialized receptors called nociceptors. A system can react to damage without feeling anything, which is the reason a withdrawal reflex alone does not settle the matter.
Drawing on more than 300 scientific studies, the authors recommended that all cephalopod molluscs and decapod crustaceans be regarded as sentient animals for the purposes of UK animal welfare law. The Animal Welfare (Sentience) Act 2022, which received assent on April 28, 2022, now defines "animal" for its purposes to include any vertebrate other than humans, any cephalopod mollusc, and any decapod crustacean.
The eight criteria and a machine
The LSE review graded the evidence for each animal group against eight criteria. The table pairs each with a short note on how it could apply to software. The notes are an illustrative reading, not a published assessment, and they show that the framework assumes a living body.
| Criterion in the review | Illustrative note on AI |
|---|---|
| Nociceptors | No biological receptors; a sensor that flags damage is not the same thing |
| Integrative brain regions | Unclear what the software equivalent would be |
| Connections between the two | Defined in anatomical terms that do not map onto code |
| Responses affected by local anesthetics or analgesics | Not applicable |
| Motivational trade-offs between threat and reward | A system can be built or trained to trade off costs, and can describe doing so in text |
| Flexible self-protective tactics after injury or threat | Robots and agents can be built to adapt, which proves little by itself |
| Associative learning beyond habituation | Met by ordinary machine learning in a technical sense, so it cannot separate feeling from non-feeling here |
| Behavior showing the animal values analgesics when injured | Not applicable |
The pattern is informative. Several criteria depend on anatomy and drugs. The rest are behaviors that machines can reproduce through training or design without any suggestion of an inner state. The framework was tuned to organisms whose behavior is evidence of feeling because evolution links the two, and that link cannot be assumed for software.
The precautionary principle for animals
On April 19, 2024, a group of scientists and philosophers published the New York Declaration on Animal Consciousness at New York University. The page lists more than 600 signatories. It says there is strong scientific support for attributing conscious experience to mammals and birds, and at least a realistic possibility of it in all vertebrates and many invertebrates, including cephalopods, decapod crustaceans, and insects. Where such a possibility exists, the declaration says it is irresponsible to ignore it in decisions affecting that animal.
The declaration concerns animals. Extending its precaution to AI would be a separate argument, because the evidence that supports it for animals (shared brain structures, shared evolutionary history, pain-related behavior) is absent or contested for machines.
What this means for claims about AI feelings
The animal record offers a method more than an answer. It asks for evidence from several independent lines, says what would count against sentience, and records confidence levels instead of declaring yes or no. The LSE authors state that low confidence about a criterion means the evidence is thin, not that sentience is unlikely. Applied to AI, that standard would reject a chatbot's statement about its feelings as sufficient, since such systems are trained on human writing about feelings, and it would also reject confident claims that no AI could ever feel.
Checklist for reading a claim that an AI is sentient
- Does the claim define sentience, and does it separate feeling from reacting to a stimulus?
- Is the evidence the AI's own words, or something measured independently of what it says?
- Does the evidence come from more than one line, as the eight-criterion review requires?
- Does the source state its confidence and what would change its mind?
- Is the claim attributed to a named researcher or body that can be checked?
FAQ
Has any government recognized an AI system as sentient?
No such recognition was found in the sources checked. The UK sentience law discussed here applies to vertebrates, cephalopods, and decapod crustaceans, and it concerns animals only.
What is the difference between sentience and consciousness?
Sentience usually refers to having feelings such as pain or pleasure, while consciousness is the broader term for subjective experience. The two are often used together in animal studies, and the New York Declaration speaks of conscious experience.
Why does the nociception distinction matter for AI?
A machine can detect damage and respond, for example a robot that retreats from heat. The LSE review treats that kind of detection as separate from feeling pain, so a protective response does not show sentience.
Can the eight criteria be used to test an AI?
They were designed for animals and are not a validated test for software. Some criteria have no machine counterpart, and others can be met by design without any inner state.
Related reading
- Can AI feel emotions
- Why emotion recognition errs
- Forecasts for machine sentience
- Whether AI can dream
More in AI Policy and Ethics.
Last verified: October 2026
Sources
- Review of the Evidence of Sentience in Cephalopod Molluscs and Decapod Crustaceans, Birch, Burn, Schnell, Browning, and Crump, London School of Economics, November 2021
- Animal Welfare (Sentience) Act 2022, UK Parliament, 2022
- New York Declaration on Animal Consciousness, New York University, April 19, 2024