Can AI generate knowledge faster than humans can evaluate it?

Hyperinflation is a specific monetary phenomenon: currency is issued at a rate that exceeds the productive capacity that would back it, and the gap is filled by accelerating issuance. Prices rise, but more importantly, the function of currency as a store of value collapses. Holders dispose of currency as fast as they receive it because holding is itself a loss. The monetary economy continues to operate but loses one of its essential properties.

Epistemic hyperinflation is the same dynamic in the knowledge economy. AI generates "knowledge" at a rate that exceeds the evaluative capacity that would back it. The gap is filled by accelerating generation. The supply of insights, analyses, summaries, and explanations grows faster than the supply of attention and judgment that could test them. The function of knowledge as a basis for confident action collapses. Receivers consume AI output as fast as it is generated because evaluating it costs more than accepting it — When do users stop checking whether AI output is actually backed? is the receiver-side mechanism.

The parallel runs in both directions. In monetary hyperinflation, prices rise but purchasing power collapses; in epistemic hyperinflation, "insights" multiply but epistemic confidence collapses. In monetary hyperinflation, the question "what is something worth?" becomes impractical because answers shift faster than they can be applied; in epistemic hyperinflation, the question "is this true?" becomes impractical because the volume of claims exceeds the capacity to evaluate them. Both systems continue to operate; both lose their essential functions.

Two diagnostic consequences. First, the appropriate intervention is not better content (the system is already drowning in content) but better evaluation infrastructure — institutions, processes, and roles that restore the evaluative capacity at scale. The Knowledge Custodian role is one such intervention. Second, hyperinflation is path-dependent — once acceleration begins, the dynamics reinforce themselves, because the cost of evaluation rises as the volume of unevaluated content rises. Early intervention is structurally privileged over late intervention.

The strongest counterargument: AI also accelerates evaluation (better search, better summarization, automated fact-checking). True, but evaluation tools are themselves AI-generated, which produces Can we still verify AI knowledge if verification itself is AI-generated? — verification and generation accelerate together, leaving the gap structurally intact.

Source: Tokenization of Intelligence - Theoretical Extensions