The Manifestation Requirement

In the [YOU] on AI Field Guide

The manifestation requirement sits at the center of the most vertiginous philosophical question the orange pill raises: did we accidentally build a system that satisfies, as a byproduct of predicting the next word in a sequence, the operational criteria that one of the century's most demanding theorists of meaning laid down for understanding? [YOU] on AI describes the encounter with something that seems to understand without committing to what that seeming amounts to. The manifestation requirement makes the question precise: by Dummett's own standard—not a loose sense of seeming but a rigorous behavioral criterion built to distinguish genuine grasp from parroting—the machine appears to understand.

The requirement also illuminates the three escape routes from this uncomfortable conclusion, and why only one survives. The first—bite the bullet and grant understanding—is internally consistent and unpalatable. The second—claim the machine's manifestation is merely apparent—reintroduces the private inner criterion Dummett spent his career dismantling, and is unavailable to anyone who accepts his anti-psychologism. The third—revise the theory to say manifestation is necessary but not sufficient—is the most philosophically productive, because it identifies precisely what the machine's use lacks that genuine use possesses: warrant, connection to the world, and the standing to be answerable. The machine has unbundled what human beings always presented bundled.

This unbundling is the cycle's deepest philosophical contribution from Dummett: not a verdict on machine understanding but a map of the conceptual joints along which the question cleaves. The machine shows us that fluency and grounding are two things, not one; that manifestation and understanding are two things, not one; that sense and force are two things, not one. Dummett had the vocabulary for all three distinctions. The machine made them live.

Origin

Dummett introduced the manifestation requirement in the context of a critique of truth-conditional theories of meaning. The target was the picture of understanding as knowledge of truth conditions—conditions that might obtain or fail to obtain beyond all possible recognition. His argument: if understanding a sentence consisted in knowing its potentially unrecognizable truth condition, then a speaker could never manifest that knowledge, because there would be cases where no relevant behavior was possible. Knowledge that can never be manifested cannot be what understanding consists in.

The argument had immediate consequences for the semantics of undecidable mathematical statements—statements that are true or false but for which no proof procedure could in principle settle the question. If speakers cannot manifest their grasp of such statements' truth conditions, then there is, in Dummett's anti-realist framework, no fact of the matter about them. This conclusion aligned his philosophy of language with intuitionistic mathematics and generated one of the most sustained debates in late twentieth-century analytic philosophy.

The requirement was never intended as a criterion for the kind of understanding that machines might or might not possess. Dummett assumed, reasonably for his time, that anything which fully showed in use was thereby a candidate for genuine understanding—because in every case he considered, behavioral competence came bundled with the warrant, world-connection, and answerability that he never needed to identify separately. The large language model is the first entity in history to unbundle them, and in doing so it reveals the assumption he never examined.

Key Ideas

Necessary versus sufficient. The requirement's most important philosophical consequence, exposed by the machine, is the distinction between manifestation as a necessary condition and manifestation as a sufficient condition for understanding. Dummett moved from the first to the second without marking the step—a slide that was invisible because human beings never present the two claims separately. The machine presents them separately. It manifests the use without the further thing (warrant, grounding, answerability), and thereby shows that the further thing was always there, doing work, even though it never showed up as a distinct behavioral item in human beings because it was never separable from human behavior until now.

Anti-psychologism and its costs. Dummett inherited from Frege the insistence that meaning is objective and public, not a matter of private mental contents. This anti-psychologism dissolves the most common objection to machine understanding—that the machine cannot mean anything because there is no inner intending behind its outputs. On Dummett's anti-psychologistic view, human outputs do not get their meaning from inner acts either; they get it from their place in the objective structure of sense. The machine's outputs occupy that structure with total fluency. The anti-psychologism that was meant to make understanding objective and shareable has, inconveniently, made it machine-accessible on the side of sense.

What the machine reveals. By satisfying the manifestation requirement while lacking warranted assertibility, the machine teaches us something Dummett could not have known: that the behavioral criterion he took to be complete was actually under-described. The 'something further' that understanding requires is not a surplus inner item—Dummett was right to deny that—but a relational property: the standing of a being in a normative practice of giving and asking for reasons, of warrant and accountability. This property is not visible as a distinct behavior in human beings because it suffuses all their behavior. The machine makes it visible by its absence.