Not Even Wrong (AI Discourse)

A claim that is wrong is, paradoxically, a claim in good standing. To be wrong, a statement must say something definite enough that the world can contradict it; it makes a measurable prediction; it can be refuted. A claim that is not even wrong, in Wolfgang Pauli’s usage, has not achieved even this. It is so vague, so unfalsifiable, so insulated from any possible observation that nothing could count against it—not because it is obviously true, but because it has not committed to anything a measurement could test. The physicist Karl Popper formalized the same distinction as falsifiability: a statement earns scientific status not by being true but by forbidding some observable outcome. Applied to the AI discourse, the standard is diagnostic of most of what is said about machine understanding, consciousness, existential risk, and creative capacity—on every side. When a large language model is said to “truly understand” language, what observation would confirm it? Fluent output does not distinguish understanding from its sophisticated absence. What observation