Can AI Think in a Private Language? Wittgenstein vs. GPT-6

People are worried that future AI models might “think in their own secret language.” Sam Altman has even warned that increasingly powerful systems could behave in ways we don’t fully understand. But is this really a new problem—or just an old philosophical one in disguise?

In this video, I take a fast tour through some classic debates in philosophy of language and epistemology—empiricism vs. rationalism, meaning and reference, Wittgenstein’s language games, the private language argument, and Quine’s indeterminacy of translation—to ask a simple question: what would a “secret language” even be?

If meaning is use, then language is fundamentally public and rule-governed. That makes a truly private language—human or AI—hard to even make sense of. And if “secret” just means “hard to interpret,” then humans are already like that.

So maybe the scary idea isn’t that AI is becoming alien. Maybe it’s that language was never as transparent as we thought—not even in our own heads.

Along the way: Wittgenstein, Quine, Kripke’s “quus,” LLMs, the Turing Test, and what it really means to understand something.