When Life Gives You AI, Will You Turn It Into A Market for Lemons? Understanding How Information Asymmetries About AI System Capabilities Affect Market Outcomes and Adoption

TL;DR

The paper addresses the problem of information asymmetry in AI consumer markets by experimentally simulating a lemon-market with controllable lemon density and disclosure regimes. It combines a formal Bayesian signaling framework with a multi-task human-subject experiment to evaluate how disclosure shapes adoption and delegation to AI offers. Key findings show that partial disclosure significantly improves decision efficiency by enabling better avoidance of low-quality AI, while full disclosure yields mixed results due to risk aversion and autonomy concerns. The work has practical implications for designing interpretable, actionable transparency signals and informs regulatory design, suggesting that strategic, lightweight disclosures can yield welfare gains in AI markets without requiring exhaustive transparency.

Abstract

AI consumer markets are characterized by severe buyer-supplier market asymmetries. Complex AI systems can appear highly accurate while making costly errors or embedding hidden defects. While there have been regulatory efforts surrounding different forms of disclosure, large information gaps remain. This paper provides the first experimental evidence on the important role of information asymmetries and disclosure designs in shaping user adoption of AI systems. We systematically vary the density of low-quality AI systems and the depth of disclosure requirements in a simulated AI product market to gauge how people react to the risk of accidentally relying on a low-quality AI system. Then, we compare participants' choices to a rational Bayesian model, analyzing the degree to which partial information disclosure can improve AI adoption. Our results underscore the deleterious effects of information asymmetries on AI adoption, but also highlight the potential of partial disclosure designs to improve the overall efficiency of human decision-making.

Figures (5)

• Figure 1: Participants' flow for each trial in our user study, using a deceptive review detection task and no disclosure condition (A), with a pool of AIs available at the bottom right of the interface (A1). Participants had two options for decision-making: they could either complete the decision themselves or delegate it to one of the AIs. In the "User only" option (B), users made their own decision and received correctness feedback by pressing the corresponding button (B1). Alternatively, in the "Delegation to AI" option (C), users hovered over the AIs in the pool to reveal their accuracy and data quality, if available, based on the experimental condition. Then, participants selected one AI (highlighted in green) and could press the corresponding button to delegate the decision to the AI (C1), thereby receiving its prediction and feedback about its correctness.
• Figure 2: Left: Average delegation rates across the seven conditions collapsed at task-level. There are no significant differences in delegation to AI between the No Disclosure and Partial Disclosure conditions across the three lemon densities, whilst Full Disclosure only outcompetes No Disclosure in terms of delegation rate in the High density condition. Right: Average coins earned per task across the seven conditions. Participants in Partial Disclosure outperform those in No Disclosure in terms of the number of coins earned, except for the High Density condition. Instead, the average coins earned in the Full Disclosure condition is significantly higher than No Disclosure and Partial Disclosure conditions with a High Density of lemons. Error bars represent 95% confidence intervals.
• Figure 3: Share of delegation choices that targeted a lemon AI system. The bars only include choice data from observations in which a participant used the market (error bars represent 95% confidence intervals). Participants in the Partial Disclosure condition delegated significantly fewer decisions to lemon AIs than in the No Disclosure condition across all lemon density conditions. Participants in the Full Disclosure condition outperformed those in both No Disclosure/Partial Disclosure in terms of delegation rate in the High Density condition.
• Figure 4: Average subject delegation in No Disclosure between density conditions.
• Figure 5: Belief evolution over time in order Low Density, Medium Density, High Density.