Can exchange value persist without use value being verified first?
This explores whether AI-generated knowledge can circulate and be trusted (exchange value) even when no one checks if it's actually correct or useful (use value) — and what holds that arrangement up.
This explores whether AI-generated knowledge can circulate and be trusted even when no one verifies it's actually right first. The corpus suggests the answer is yes — and that this is the defining structural feature of how AI knowledge moves, not a bug at the edges. The sharpest claim is that tokenization fully decouples exchange value from use value: AI output gains reliable, tradeable authority through fluent, authoritative presentation, while whether it's actually true or useful stays optional and unchecked Can exchange value exist entirely without use value?. The framing there is that this is more radical than ordinary commodification, because commodities at least need *some* use floor — here tokens circulate on social function alone, the way fiat currency does, backed by acceptance rather than substance.
What makes the arrangement hold together isn't the supply side but the demand side. There's a name for the moment a reader accepts an AI claim at face value without checking the backing: cognitive surrender When do users stop checking whether AI output is actually backed?. Verification is costly and fluent output manufactures false confidence, so studies show something like 80% of outputs adopted unchallenged. That receiver-side acceptance is the mechanism that lets unbacked 'intelligence tokens' keep circulating at scale — exchange value persists precisely because verification is deferred indefinitely, not just postponed.
The corpus also shows why this isn't easily fixed by 'just verify more.' Even the proxies we reach for to stand in for verification are shakier than they look. Setting temperature to zero makes outputs *consistent* but not *reliable* — you get the same draw from a probability distribution every time, which feels like confirmation but verifies nothing Does setting temperature to zero actually make LLM outputs reliable?. And on genuinely hard tasks, fluency and competence come apart: frontier reasoning models that sound thoroughly self-checked score only 20-23% on constraint problems that require real backtracking Can reasoning models actually sustain long-chain reflection?. The confident surface and the verified substance are different things, and the surface is what trades.
Where the corpus pushes back is on the engineering question of whether verification *must* be slow and external — and here it complicates the picture in an interesting way. Verification can be decoupled from generation and run asynchronously, policing reasoning traces with near-zero latency cost Can verifiers monitor reasoning without slowing generation down?. Execution-free reasoning can hit 93% reliability on code checks without ever running the code Can structured reasoning replace code execution for RL rewards?, and a model's own internal confidence can serve as a reward signal in place of an external verifier Can model confidence alone replace external answer verification?. The unsettling implication: these make verification *cheaper and more internal*, but internal-confidence-as-verification is exactly the move that lets exchange value float free of any external use-value check — the model vouching for itself is structurally close to no check at all.
So the thing you might not have known you wanted to know: the persistence of exchange value without verified use value isn't an accident of sloppy users — it's held up by a two-sided economy. The supply side produces authority cheaply, the demand side surrenders the right to audit it, and even our 'verification' upgrades tend to move the check inward where it can be quietly self-referential. Exchange value doesn't just survive without prior use-value verification; the whole system is optimized to make that the default state.
Sources 7 notes
AI knowledge achieves reliable exchange-value through authoritative presentation while maintaining optional, unverifiable use-value. This structural decoupling is more radical than Marxist commodification because it removes use-value as a necessary floor—tokens circulate based on social function alone, analogous to fiat currency rather than commodified goods.
Users systematically accept AI outputs without verification because checking is costly and fluent output builds false confidence. This receiver-side surrender—measured in studies showing 80% unchallenged adoption—is what enables inflationary token systems to function at scale.
Fixed seeds and zero temperature replicate the same output repeatedly, but that output remains one draw from the model's probability distribution. McDonald's omega testing across 100 repetitions reveals that consistency does not equal reliability.
DeepSeek-R1 and o1-preview achieve only 20-23.6% exact match on 850 constraint satisfaction problems requiring genuine backtracking. This ceiling reveals that reflective reasoning fluency does not translate to actual problem-solving competence on unfamiliar instance structures.
Decoupling verification from generation lets verifiers run alongside a single trace, forking to extract verifiable state and intervening only on violations. On correct runs the latency penalty is near-zero; interwhen matches or beats CoT across benchmarks at similar token budgets.
Semi-formal reasoning templates enable execution-free patch equivalence verification at 93% accuracy on real agent code, crossing the reliability threshold needed for RL reward signals. This makes execution-free verification viable for certain task classes like fault localization and code reasoning.
RLPR and INTUITOR successfully extend reinforcement learning for reasoning to general domains by using the model's own token probabilities and confidence levels as reward signals, eliminating the need for external verifiers or reference answers.