The Philosophical Investigations, published two years after Wittgenstein's death, replaces the austere logical architecture of the Tractatus with a method: look and see. Words do not get their meaning from correspondence with objects or from shared essences. They get their meaning from use — from their role in language games played by specific people in specific contexts for specific purposes. The book proceeds through numbered remarks rather than sustained argument, inviting the reader to examine cases and dissolve philosophical confusions rather than solve them. Its central concepts — language games, family resemblances, forms of life, the private language argument, the beetle in the box, and rule-following considerations — reshaped twentieth-century philosophy of language and mind and now provide the sharpest available framework for thinking about AI.
The Investigations was composed over roughly two decades of lecturing and note-taking at Cambridge. Wittgenstein revised it obsessively and never finished it to his satisfaction. The published text assembles his late manuscripts, with the second part later contested on textual grounds. The method is deliberately non-systematic: the reader is invited into specific examples and asked to notice what is actually there, not to accept a theory that replaces the one being criticized.
Where the Tractatus sought the essence of language, the Investigations denies that language has an essence. The word game does not name a single property shared by all games; it picks out a network of overlapping similarities — family resemblances. This dissolves the classical theory of meaning that had dominated Western philosophy since Plato and shaped the dream of perfect language through Leibniz, Frege, and the Tractarian Wittgenstein himself.
The book's importance for the AI moment is specific. The natural language interface arrived without any serious philosophical account of what it is for a machine to participate in language games it did not develop. The Investigations supplies that account, or at least the conceptual apparatus required to formulate the question with precision. When Edo Segal describes the feeling of being met by Claude, Wittgenstein's framework makes the description philosophically exact: to feel met is to have one's language game recognized and participated in.
The book's method also models a discipline the AI moment demands. Don't think, but look. Resist the theoretical gesture that rushes to resolve the question of whether machines understand. Examine what actually happens in the interaction. The resolution, in Wittgenstein's framework, is often that the question was malformed — a confusion generated by the grammar of the concepts we are using rather than by a genuine fact about the world.
Written in fragments across the 1930s and 1940s, revised continuously, and left unfinished at Wittgenstein's death in 1951. Published by his literary executors G.E.M. Anscombe and Rush Rhees in 1953 with parallel German and English text. The later Wittgensteinian tradition — Anscombe, Kripke, Stanley Cavell, Cora Diamond, Charles Taylor, Hubert Dreyfus — developed its implications across multiple disciplines.
Meaning is use. The meaning of a word is its role in a language game — the practice within which it is employed.
Family resemblances. Concepts pick out networks of overlapping similarities rather than shared essences. There is no single property all games share.
Forms of life. Language games are embedded in forms of life — the background of shared practices and reactions that makes linguistic meaning possible.
Rule-following is a practice. Applying a rule correctly is not determined by the rule itself but by participation in a community of practitioners who agree in judgments.
Philosophy as therapy. Philosophical problems are not solved but dissolved — shown to be confusions generated by misapplications of ordinary concepts.
Whether the Investigations offers a positive theory of meaning or only a therapeutic dissolution of pseudo-problems is among the oldest debates in Wittgenstein scholarship. Saul Kripke's 1982 reading — Wittgenstein on Rules and Private Language — argued the book contains a devastating skeptical argument; Baker and Hacker countered that the argument is descriptive rather than skeptical. Both readings have consequences for how Wittgenstein's framework applies to AI.