Wittgenstein
Meaning Performed
The classic idea
Meaning is use. A word doesn't point to a fixed inner essence, its meaning is constituted by how it functions inside a "language game," a shared practice among people.
The AI inversion
Language models become a live test of the limits of understanding through use, producing fluent, contextually appropriate language without anyone being able to confirm whether anything resembling understanding sits behind it.
"AI shows how far language can go without anyone knowing if it understands."
Wittgenstein's later philosophy let go of the idea that meaning lives inside words like a hidden essence waiting to be found. Meaning, he argued, is what a word does inside a practice, a "language game." You understand "checkmate" by playing chess, not by inspecting the word. Use is the whole story.
A language model is, in a strange way, the purest possible test of that claim. It has never played a single human language game in the way Wittgenstein meant, shared form of life, embodied practice, stakes. And yet it produces language that fits use after use, context after context, with startling fluency. If meaning really is just use, the model is doing something close to meaning things.
That's the discomfort. Wittgenstein gave us a theory built to explain human meaning-making, and a machine came along that satisfies the theory's behavioral test while clearly lacking the form of life the theory assumed would be underneath it. Either the test was incomplete, or "understanding" was always more behavioral than we wanted to admit.
Why it matters
If fluent, contextually correct use is all meaning ever required, we may need a new word for whatever AI is still missing.
The question AI brings back
Is there a private, felt understanding behind use at all, or was that always a story we told about ourselves?
Sources for further reading
- Ludwig Wittgenstein, Philosophical Investigations
- Hubert Dreyfus, Being-in-the-World
