How does AI writing escape the conversations that govern knowledge?

In human knowledge production, claims are raised as moves in a conversation — addressed to an audience, raised as responses, concurrences, or objections to existing claims. The conversation is the governing mechanism: it decides which claims circulate, which compound, which are discarded. Claims are embedded in social, cultural, and economic production, which is why knowledge has any reliability at all.

AI-generated claims are produced outside this conversation. They are not responses. They are not addressed to anyone in particular. They do not take up a position relative to other positions, because the system that generates them has no position. The text appears as a supplement to discourse — adjacent to it, but not participating in it. Because it does not participate, it is not governed. The ordinary mechanisms that filter, credit, and refine knowledge claims cannot act on it.

The result is inflation in the monetary sense: a proliferation of tokens (claims) disconnected from the backing (the conversational work) that would normally give them value. Because Does AI generate diverse claims or diverse perspectives?, the apparent diversity of output masks a collapse of the conditions under which knowledge becomes reliable. This is not about hallucination or factual error — it is about the structural dislocation of claims from their governing context.

The strongest counterargument: conversations eventually absorb AI-generated claims and govern them ex post. But the volume is the problem. Governing mechanisms cannot scale to a stream of disembedded claims that never enter conversation in the first place.