The Last Costly Signal: How Generative AI Collapses Competence Signaling and Why Liability Sustains Markets for Expert Services
arXiv:2607.26327
The paper models how generative AI makes expertise reports cheap, breaking traditional costly‑signal mechanisms in expert‑service markets, and shows that liability‑based warranties can restore reliable signaling.
Abstract
Generative artificial intelligence has reduced the cost of producing convincing artifacts of expertise-reports, analyses, proposals-to nearly zero. Signaling theory predicts that signals whose informational content rests on production cost lose that content when production becomes cheap. We formalize this prediction for markets for expert services, a class of credence goods, by modeling generative AI as a compression of the discernible headroom between what machines produce at negligible cost and what buyers can distinguish at all. Below a critical headroom, no separating equilibrium in production-side signals exists; the market pools, high-competence providers earn no premium, and those with outside options exit-Akerlof's lemons dynamic. We show that an outcome-contingent signal-a warranty backed by damages D with ex-post verifiability phi-restores full separation for any level of AI capability whenever phi*D >= v, where v is the value of a solved problem. The expected cost of liability depends on whether the problem is solved, not on document production costs. A corollary shows that provenance certification (e.g., C2PA), whose cost is type-independent, cannot restore separation. Agent-based Monte-Carlo simulations illustrate the dynamics. Two further results endogenize contract institutions: civil procedure costs set a minimum ticket size v_min below which no credible enforcement threat exists; under liability insurance, separation depends on retained risk or risk-rated premiums. We state falsification conditions and propose a preregistered choice-based conjoint experiment with decision-makers in the German-speaking B2B expert-services market.
26 pages, 7 figures. JEL codes: D82, D86, L15, L84, M31, O33. Simulation code and numerical results available at: https://doi.org/10.6084/m9.figshare.33106946