Episode 13 — Wittgenstein’s Shadow

When rules failed, language began to live.

Early AI was built on a powerful assumption: that intelligence could be reduced to rules—formal steps a machine could follow. But language refused to cooperate.

Episode 13 turns to Ludwig Wittgenstein, whose later philosophy challenged the idea that meaning is fixed by definitions or logical form. Instead, meaning arises from use—from the ways words function inside human practices, contexts, and “forms of life.”

This episode explores why rule-based AI (GOFAI) struggled with real language, and why modern AI shifted toward data-driven models that learn patterns of use rather than executing hand-crafted rules.

Reflection

Wittgenstein’s early work (the Tractatus) helped sustain a modern dream: that language mirrors reality through a logical structure, and that meaning can be captured in a formal system. It is not surprising that early AI inherited that dream. If language is fundamentally rule-governed, then building intelligent machines would look like building the right rulebook.

But Wittgenstein later reversed direction. In Philosophical Investigations, he argued that language is not a mirror but a human activity. Words do not carry meaning like labels attached to objects. They gain meaning through use—in shared practices, social contexts, and shifting circumstances. Language is not one system; it is many “language-games.”

This casts a new light on why rule-based AI faltered.

A machine can follow rules with perfect discipline.
But real language is not just rule-following. It is flexible, context-sensitive, and often underdetermined by grammar alone. The same sentence can function as a warning, a joke, a promise, or a threat—depending on the situation.

Modern AI took a different route: instead of trying to encode meaning as rules, it learned from patterns of use in large bodies of human text. In that sense, contemporary language models align more closely with Wittgenstein’s later insight: meaning is not extracted from essence—it is inferred from practice.

Still, an important tension remains:

AI can approximate patterns of use.
But it does not participate in the human form of life that gives those patterns their depth.

So Episode 13 leaves us with a careful question—not only about machines, but about ourselves:

If meaning is use, what kind of “use” counts as understanding?

Eisode 12: Birth of Artificial Intelligence / All Episodes / Episode 14 Architecture of Truth (coming soon)

Questions for Thinking

  1. Wittgenstein suggests that meaning comes from use, not from fixed definitions.
    Where in your own life have you seen a word’s meaning shift dramatically with context?

  2. Rule-based AI struggled because language is messy and situational.
    What do you think rules are good for—and where do they inevitably fall short?

  3. If a system can generate language that “fits” a context, does it understand—or does it only perform?
    What would make the difference?

  4. Modern AI learns from patterns in human language use.
    What does that help it capture—and what might it still miss because it does not live a human life?

  5. If meaning depends on shared practices, what responsibility do we have for the language-games we create—online, at work, and in public life?