Words With Blurry Edges
A series on the vocabulary of mind — five words we use to describe consciousness, traced through the long history of their shifts.
In 2022 the United Kingdom passed a law about sentience without defining it. The Animal Welfare (Sentience) Act created a committee, gave it statutory standing to scrutinize government policy, and left the word at its center to mean whatever the reader already assumed it meant. A review commissioned alongside the act had to supply the working criteria the statute itself left out. That same year, a Google engineer went public claiming that a chatbot he had been testing was sentient — and meant something almost entirely unrelated by the word.
Ask three researchers across animal welfare, machine learning, and philosophy of mind what it means to be sentient, and you will get three answers with almost no overlap. Ask again in ten years and all three will have moved. This is language doing what language does on hard problems, stretching a single word across a question that hasn't settled yet, and likely won't anytime soon.
There's a cluster of words that orbit the idea of mind: sentience, attention, intelligence, empathy, sapience. They're in heavy use right now in debates about AI, animal cognition, moral status, and consciousness. Behind each one sits a root in Latin or Greek, centuries of competing technical definitions, folk meanings that crept in and hardened, and recent redefinitions pulled in several directions at once by various subjectively technical fields. Each is doing double or triple duty between fields that don't communicate or read each other's work, with every field confident it has the term right.
This series is a look at each word across time rather than a snapshot. It's not a disambiguation guide, the ambiguity's the point. When a word's meaning shifts, it's usually because something underneath it shifted first — a change in what people thought they were describing. I'm interested in when each word changed, what was changing around it when it did, and what the competing definitions reveal about the assumptions of the people using them. Less "here's the right definition" and more an account of where each word came apart (and when).
The disarray is evident, if you know where to look. In The Edge of Sentience (2024), philosopher Jonathan Birch separates out several incompatible senses of "sentience" in circulation across philosophy, science, and policy. He notes how much confusion the overlap causes, enough that serious work in the area now tends to open with hedges and caveats. The other 4 words are similar.
Why these five? At the moment, all five are contested. The contending fields (AI, animal cognition, philosophy of mind) have each landed on a different meaning.
Sentience now means two incompatible things. To philosophers of mind, it points at phenomenal consciousness and the hard problem: the claim that there is "something it is like" to be a given system. The phrase is usually credited to Thomas Nagel's 1974 bat essay. Though B. A. Farrell had asked "what would it be like to be a bat?" in Mind in 1950, Nagel made it stick. In utilitarian ethics and animal-welfare law, sentience came to mean something narrower: the capacity for experience that feels good or bad (to suffer, to enjoy) regardless of intellect. That lineage runs from Bentham's "Can they suffer?" in 1789 to Peter Singer making it the criterion for moral status in 1975. Buddhist usage, meanwhile, treats "sentient beings" as a near-universal category, extending to animals and insects, though the exact boundary is itself debated within the tradition. These are different theories of what matters wearing the same costume.
Attention was a phenomenological term first: what it feels like to direct the mind at something. William James in 1890 treated it as introspectively obvious, defined by selection and withdrawal. Then it became an information-processing construct: Donald Broadbent's 1958 filter model, Daniel Kahneman's 1973 account of attention as a single limited resource spread across tasks. Every version so far assumed a bounded subject who couldn't attend to everything; attention was defined by what it left out. Then in 2017 the transformer architecture attached "attention" to a specific mathematical operation, a weighted sum over representations, with no bounded subject and no limit: it compares every element to every other. Within a few years, "attention" in technical usage meant the mechanism first and the mental phenomenon second. The constraint that defined the concept is gone, the word remains.
Intelligence became quantifiable in 1904–05. Charles Spearman proposed that a single latent factor — g — sat behind the correlations among test scores; Alfred Binet and Théodore Simon built the first practical test, meant to identify schoolchildren who needed extra help. (Binet warned against treating the result as a fixed verdict, which is exactly what happened.) A word that had previously meant a faculty, the capacity to grasp general truths, was reworked into a number on a scale. A third sense arrived in 1955, when John McCarthy coined "artificial intelligence" for the Dartmouth proposal, picking a fresh term not tied to psychometrics or to "cybernetics." The word now has three main prongs: the psychometric g, folk "smartness," and AI's working sense of "can do things we thought needed a person" (which moves every time the technology does).
Empathy is the youngest of the five, its referent changing almost at the moment of coinage. Edward Titchener minted the English word in 1909 to translate the German Einfühlung, "feeling-into" (at the time a term of art in aesthetics from Robert Vischer's 1873 work on projecting human feeling into visual form, later extended by Theodor Lipps to the perception of other minds). A generation later, "empathy" had shifted to mean feeling into other people. By mid-century, it meant understanding and sharing someone else's emotional state. A reader today hears none of these aesthetics, and the word is still unsettled: today's arguments over affective versus cognitive empathy (or over whether empathy is even a reliable moral guide) are the same instability it was born with.
Sapience is the quiet case. Latin sapere meant both "to taste" and "to be wise" — the root of sapor, flavor, and of Homo sapiens, the name Linnaeus gave us in 1758: the wise one, not the intelligent one. It also echoes the sophia of philosophy. Sapience has always pointed at a capacity to discern, to judge well, to know what matters; wisdom, as something distinct from cleverness. That distinction keeps getting absorbed into "intelligence" and subsequently mourned as lost. You can watch it in mid-century science fiction where "sentient" became the generic label for a smart alien and "sapient" got pressed into service for "human-level reasoning", at which point it just means very intelligent. No field is fighting over sapience, it simply lost the distinction without a fight.
Now, I hope those brief summaries gave you a glimpse of what's in store! The plan is one long-form entry per word, following each term through its history without forcing a verdict about what it "really" means. The goal isn't to resolve the ambiguity, it's to get a better map of each word's grounding so that terms like "Artificial General Intelligence", "Empath", and "Proto-Consciousness" read as places where several older disagreements have collected under one word.
Entry one is sentience: the most concrete at its origin (bodily feeling and nothing more) and the one whose meaning has changed most since. It's also the word doing the most legal and ethical work right now, which is where the blur starts to cost something. See you soon!