Informational Opportunism

In the AI Story

The concept synthesizes Williamson's opportunism framework with Han's aesthetics of the smooth and information economics' work on asymmetric information (Akerlof's lemons, Spence's signaling). Traditional opportunism involves one party exploiting superior information—the used car seller knows the car is defective, the buyer does not, and the seller strategically conceals the defect. AI reverses the pattern: the system producing the output may have no 'knowledge' of quality in any meaningful sense, and the human receiving it may lack the expertise to assess it, creating an informational vacuum where surface characteristics become the de facto quality signal despite being decorrelated from actual quality. The strategic dimension enters when workers or organizations accept AI output without verification, implicitly trading present productivity for future risk, knowing that surface inspection is inadequate but proceeding anyway because depth evaluation is expensive.

The smooth surface is not incidental but structural: AI systems are optimized to produce fluent, well-formatted, confident-sounding output regardless of underlying soundness. Temperature settings can adjust creativity/randomness, but they do not reliably correlate with accuracy. An AI can produce a confident, polished, entirely fabricated legal case citation with the same surface quality as a correct one. The fabrication is not intended deception—the system has no intent. But the effect is opportunistic from the receiver's perspective: the smooth presentation induces acceptance, the acceptance decision is based on inadequate information, and the costs of the error (when the fabricated citation is discovered during litigation) fall on the party who accepted without verifying. This is informational opportunism in structure if not in intent—one party capturing benefits while externalizing costs made possible by informational asymmetry.

Governance requires costly verification mechanisms that the productivity optimization regime systematically eliminates. Reference checking: every claim made by AI should be traceable to a primary source the evaluator has actually examined—expensive in time, resistant to automation. Logical auditing: outputs should be evaluated not for surface coherence but for the validity of underlying reasoning—requiring domain expertise and sustained attention. Adversarial review: having a second party attempt to break the output, find its failure modes, stress-test its assumptions—doubling evaluation costs. Comparative generation: producing multiple alternative outputs for the same specification and comparing them to identify where consistency breaks down—further multiplying costs. Each mechanism addresses the informational opportunism hazard by raising the transaction cost of accepting low-quality output dressed in high-quality presentation. The costs are justified when the prevented failures exceed the evaluation expense—but the calculation is made by organizations under competitive pressure to optimize for speed, and the systematic result is depth governance underinvestment.

Origin

The term originates in this volume, extending Williamson's opportunism taxonomy to address a governance challenge his framework anticipated but did not encounter: technologies producing outputs whose surface characteristics are uninformative about underlying quality. The closest historical precedent is the medieval manuscript market, where scribal copying introduced errors that accumulating copying propagated—but even there, rough surfaces (inconsistent hand, visible corrections, marginalia) provided some information about copy quality. AI eliminates surface variation entirely, creating an informational environment where every output, regardless of provenance or quality, presents the same polished appearance. This is historically unprecedented and requires governance mechanisms that previous informational asymmetry problems did not demand.

Key Ideas

Smooth surfaces conceal quality variation. AI produces uniformly polished output regardless of underlying soundness, decorrelating surface appearance from depth quality.

Neither party may know quality. The novel structure: not one party exploiting superior information but both parties facing informational deficit relative to the output.

Accepting without verification is opportunistic. Taking the productivity gain while externalizing verification costs to the future is strategic behavior exploiting the informational gap.

Depth governance is the response. Verification mechanisms, reference checking, logical auditing, adversarial review—costly processes that surface governance systematically skips.

The hazard scales with velocity. When AI produces output faster than humans can evaluate deeply, informational opportunism compounds—each unverified output increases the risk accumulation.