Sentience
The first of five words with blurry edges. Sentience started as plain feeling, and now it turns up in UK law, AI debates, and philosophy of mind, meaning something different every time.
When the UK government introduced its Animal Welfare (Sentience) Bill in 2021, the animals it covered were vertebrates and only vertebrates. Then the Department for Environment, Food & Rural Affairs commissioned a review of the evidence for sentience in octopuses, squid, crabs, and lobsters, and a team led by the philosopher Jonathan Birch spent months working through more than three hundred studies. The review defined its subject as "the capacity to have feelings, such as feelings of pain, pleasure, hunger, thirst, warmth, joy, comfort and excitement," and concluded the answer was yes.[]Birch, Burn, Schnell, Browning & Crump, Review of the Evidence of Sentience in Cephalopod Molluscs and Decapod Crustaceans (LSE, commissioned by Defra, Nov 2021).Open source ↗ The bill was amended. When the Act passed in 2022, it set up an Animal Sentience Committee, and section 5 named the animals covered: vertebrates, cephalopodsOctopuses, squid, cuttlefish and nautiluses., and decapodsCrustaceans with ten legs: crabs, lobsters, crayfish, shrimp..
What it never did was say what sentience is. Section 5 only defines animal.[]Animal Welfare (Sentience) Act 2022, s. 5(1): 'animal' means any vertebrate other than homo sapiens, any cephalopod mollusc, and any decapod crustacean.Open source ↗ For the capacity itself you have to go to the review (a commissioned report, not law).
Two months later, in June, a Google engineer named Blake Lemoine told the Washington Post that a chatbot he had been testing (LaMDA) was sentient. Google put him on leave and said the evidence did not support his claims.[]Nitasha Tiku, 'The Google engineer who thinks the company's AI has come to life,' The Washington Post, 11 June 2022.Open source ↗ As far as I can tell Lemoine wasn't asking whether LaMDA could suffer, which is the question the UK review had been answering about crabs. He told the paper, more or less, that he knew a person when he talked to one. Same word and same year, but the two uses had almost nothing in common.
Let's start with how plain the word began. Latin sentire means "to feel." As is the case with many fusionalA language type where a single word form fuses several meanings (case, number, tense) into one inseparable piece. Latin is the standard example. languages, their words contain many meanings that we've parsed into corresponding child words and concepts in modern english: to perceive through the senses and to hold an opinion (it's the root of sense, sentiment, consent, dissent, and sentence in the sense of a judgment). Sentient, the adjective, shows up in English in the 1630s: "capable of feeling," "having the power of sense-perception." The noun sentience arrives well after, in 1817 ("faculty of sense; feeling, consciousness, susceptibility to sensation").[]Online Etymology Dictionary, 'sentient' (1630s) and 'sentience' (1817).Open source ↗ That definition lists sensation, feeling, and consciousness together as a single thing (and in everyday English they mostly are). "Feeling" can mean registering a signal, or being in a state that's good or bad, or simply being aware of anything at all. For two centuries nobody had a reason to separate them.
Then... from two points of origin people did. And as language does it branched.
The first branch: sentience as the capacity to suffer. In 1789, in a footnote to An Introduction to the Principles of Morals and Legislation, Jeremy Bentham took up the question of where to draw the line around who counts morally. His answer: "The question is not, Can they reason? nor, Can they talk? but, Can they suffer?"[]Jeremy Bentham, An Introduction to the Principles of Morals and Legislation (1789), ch. XVII, footnote. Note what he's NOT doing: He isn't asking whether animals have minds in any rich sense. Instead, he's looking for a floor: the minimum an entity needs for what happens to it to matter to it. Intellect is explicitly irrelevant.
Peter Singer picked this up in Animal Liberation (1975) and made it the foundation of a movement, using "sentience" as shorthand for the capacity to suffer and to experience enjoyment, and treating it as the single threshold for moral consideration. From there it passed into welfare science and then into law. This is the sense the Birch review used: valencedFelt as good or bad. Pain and pleasure have valence; a neutral perception of the color red might not. experience, meaning experience that feels good or bad to whoever has it. Here it's a narrow property, used to make ethical decisions. You test for it with behavior and anatomy (does the animal trade off something it wants to avoid a noxious stimulusSomething harmful or damaging to tissue: extreme heat, an electric shock, a corrosive chemical.? does it respond to analgesicsPainkillers. A change in behavior after a dose is evidence the animal was in pain.?), and a yes changes how the animal can legally be treated.
The second branch: sentience as there being something it is like. Philosophers of mind came at it from a different direction. In the late 1950s Herbert Feigl listed "sentience, sapience, and selfhood" as the three puzzles of mind, and by sentience he meant raw sensation: the qualitative feel of things.[]Herbert Feigl, 'The Mental and the Physical,' Minnesota Studies in the Philosophy of Science, vol. 2 (1958). In 1974 Thomas Nagel gave that its famous formulation: an organism has conscious experience if and only if "there is something it is like to be" that organism.[]Thomas Nagel, 'What Is It Like to Be a Bat?', The Philosophical Review 83(4), 1974, pp. 435-450.Open source ↗ His example was the bat. (The phrase isn't strictly his; B. A. Farrell had asked "what would it be like to be a bat?" in Mind in 1950[]B. A. Farrell, 'Experience,' Mind 59, 1950, pp. 170-198. but Nagel's essay is the one everybody cites.) In 1995 David Chalmers sharpened it into the "hard problemChalmers's name for the question of why any physical process feels like something from the inside, as opposed to the easier question of how the brain processes information.," and sentience, in this register, became a metaphysicalAbout what exists and what reality is fundamentally like rather than what can be measured. question: why is any physical process accompanied by experience at all?[]David Chalmers, 'Facing Up to the Problem of Consciousness,' Journal of Consciousness Studies 2(3), 1995, pp. 200-219.
One thing is missing from that formula that is present in the first branch: Nothing in "something it is like" requires the something to be pleasant or unpleasant. Imagine a system that has a vivid experience of red but no preference about it, no pleasure or distaste, and nothing at stake. On Nagel's definition, that's sentient. On Bentham's, it isn't clear whether or not it counts. Birch is explicit about this in The Edge of Sentience (2024): he separates a broad sense (any conscious experience) from a narrow one (valenced experience), argues the narrow one is the one that matters ethically, and notes how much confusion comes from people sliding between them.[]Jonathan Birch, The Edge of Sentience: Risk and Precaution in Humans, Other Animals, and AI (Oxford University Press, 2024), ch. 2, 'The Concept of Sentience'.Open source ↗ His book opens its treatment of the concept with exactly these caveats, as does most relevant contemporary work on philosophy of mind.
Two other branches are worth noting (if not writing a whole other article for)
Buddhist usage translates sattva (Pali satta, Tibetan sems can, "mind-possessing") as "sentient being," a role close to Bentham's threshold, though far older and wider. A sentient being is one that experiences suffering and pleasure, is caught in the cycle of rebirth, and is the object of compassion; the first precept and the bodhisattvaIn Mahayana Buddhism a person who vows to reach awakening for the sake of all beings. vow are both framed around the category.[]Encyclopedia of Buddhism, 'Sentient beings' (an encyclopedia entry, not a primary text).Open source ↗ It extends to animals and insects, though where exactly the boundary falls (plants, for instance) is debated within the tradition and I'd be wary of anyone who gives you a clean answer. I'm writing this article from encyclopedia-level sources here rather than primary texts, and this particular branch deserves better. For now the shape is what counts: a near-universal inclusion criterion that's generous by default and tied to being harmable.
And then AI, where the definition inverts. For animals sentience is a floor, a bare minimum we keep finding in more places than we expected: fish, then octopuses, then crabs. In the AI debate it works as a ceiling, a rare property that needs to be demonstrated before anyone takes it seriously. The 2023 report by Patrick Butlin, Robert Long, and colleagues[]Butlin, Long et al., 'Consciousness in Artificial Intelligence: Insights from the Science of Consciousness,' arXiv:2308.08708 (2023).Open source ↗ (note that it's titled "Consciousness in Artificial Intelligence," not "Sentience in") surveys the leading scientific theories of consciousness and derives indicator properties from each. It concludes that no current system looks like a strong candidate but that there are no obvious technical barriers to building one that does. So the term shifts mid-argument: the report is about consciousness, the headlines say sentience, and the ethical question (if Birch is right) is about valence: a different property again.
The AI case is awkward for the UK method, too. The animal criteria rely on evidence that's costly to fake (nociceptorsNerve endings that respond to damage or the threat of it. They detect harm; whether that comes with felt pain is the open question., wound-guardingProtecting an injured part (shielding it, favoring it, rubbing it). Treated as a behavioral sign of pain., and a crab giving up a shelter to avoid a shock). A language model's fluent output is the cheapest evidence there is to produce (it was trained on millions of people describing their inner lives). Birch calls this the gaming problem (a system can pass the tests just by being good at sounding like us), and Lemoine's chat logs are a clean example.
Each group is using the word for a different purpose.
For the ethicist and the legislator, sentience is a threshold for moral claims. The question is who can be wronged, and the word suits them because you can test for it: look for evidence, argue over it, and write it into a statute (even one that never defines it). The cost is that it has to be narrow. A capacity for suffering is a modest thing to claim for a crab, which is partly why the claim is believable.
For the philosopher of mind, it's a problem to be explained. They use it for what physical science can't yet explain. They want the opposite: the broader and more basic the term, the better it works, and whether anything is at stake for the experiencer is beside the point.
For the Buddhist tradition, it's grounds for compassion: a way of saying who is inside the circle (and the answer is nearly everyone).
For the AI debate, it's some mix of a risk to be managed and a claim to be verified (which depends on who you ask).
So. These are less rival definitions of one property than different views of what matters (same word, different stakes). A philosopher who says a thermostat-sized system "might be sentient" and a welfare scientist who says a crab "is sentient" are making claims that aren't in tension, because they aren't about the same thing, and neither would be moved by the other's evidence. The arguments that look like disputes over facts ("is it really sentient?") are often really disagreements about what the word is being used for.
I don't think the right response is to pick a winner, though it's tempting. Birch's narrow definition is the most useful one for policy, and I'd defend it for that purpose. But it doesn't describe the word's history, and a definition that works for one purpose doesn't make the others wrong about theirs.
Back to the 1817 entry: sensing, feeling, consciousness, three things in one clause. Now one word is covering all of them at once, and each field has a different piece behind it.
Next up is attention, a word with a nearly opposite problem. Sentience got stretched across too many questions. Attention stayed attention through a sudden change (the bounded mind it was named for, one that can't take everything in, isn't part of the new meaning). It's a clear case of a word outliving its own concept, and it happened in a single year (2017).