J.L. Austin

When a language model writes “I recommend you stop taking that medication,” it has not described a recommendation; it has recommended. When it writes “I promise this code will compile,” it has—apparently—promised. John Langshaw Austin is the philosopher who saw, decades before any such machine existed, that saying is a species of doing, and who built the conceptual apparatus for asking not “is this sentence true?” but “did this act come off?” His 1955 Harvard lectures, published posthumously as How to Do Things with Words, established speech act theory—the discovery that utterances perform actions (promises, assertions, warnings, apologies) rather than merely stating facts—and classified the conditions under which those actions succeed or fail. That taxonomy, the doctrine of felicity conditions, maps onto the distinctive failure modes of AI speech with eerie precision: the machine performs assertion with perfect surface form while systematically lacking the inner states, the standing, and the communal anchoring that make assertion a