Plain-language explanations nudge Europeans toward greater acceptance of algorithmic decision systems, raising willingness to delegate mainly by improving attitudes; the boost is modest and does not erase persistent privacy worries.

Main Finding

Providing a short, non-technical explanation of how an algorithmic/AI system works produces measurable but modest increases in public acceptance — primarily by improving evaluative attitudes toward the system. Explanations shift willingness to delegate decisions in the intended (more accepting) direction, but they do not eliminate persistent concerns, notably about privacy.

Key Points

• Experiment: Large-scale, randomized survey experiment with nationally representative adult samples in France, Germany, and Italy.
• Domains tested: six algorithmic/AI applications per respondent across finance, health, public services, employment, online commerce, and digital media.
• Treatment: adding a non-technical explanation to an otherwise neutral system description.
• Outcomes measured: beliefs about the system, evaluative attitudes (legitimacy/trust), and behavioral intention (willingness to delegate decisions).
• Effect size: increases in willingness to delegate are detectable and statistically reliable but modest in magnitude.
• Mechanism: mediation/decomposition shows most of the treatment effect operates indirectly via improved attitudes toward the systems; direct effects and changes in factual beliefs are smaller contributors.
• Limits: explanations do not remove core concerns — privacy worries remain salient and are relatively resistant to change.
• Robustness: effects are consistent across the three countries and across the set of domains (though domain-specific heterogeneity exists).

Data & Methods

• Sample: nationally representative adult samples in France, Germany, and Italy (large-scale; exact Ns reported in paper).
• Design: within-respondent experimental design where each respondent evaluates six systems; randomized assignment of explanation vs neutral description for each evaluation.
• Treatment content: concise, non-technical explanations of system functioning (designed to meet the kind of disclosure mandated by EU acts).
• Outcomes:
  • Beliefs (perceived accuracy, fairness, controllability, privacy risks),
  • Evaluative attitudes (trust, legitimacy, perceived appropriateness),
  • Behavioral intention (willingness to delegate decisions to the system).
• Analysis:
  • Estimation of average treatment effects on outcomes,
  • Mechanism-consistent decomposition (mediation-style analysis) to partition total effects into components mediated by attitudes and beliefs versus direct effects.
  • Heterogeneity analyses by country and domain to assess robustness.
• Limitations noted by authors: modest effect sizes, persistence of privacy concerns, and the controlled survey context (vs real-world adoption behavior).

Implications for AI Economics

• Demand-side effects: Explanations increase stated willingness to delegate and improve attitudes, implying that disclosure/ explainability can raise market acceptance and potentially adoption rates for AI-enabled services — but only modestly.
• Investment decisions: Firms and platform operators may obtain positive return on investments in understandable explanations as a trust-building tool, though such investments are unlikely alone to unlock large increases in uptake, particularly where privacy risk is perceived high.
• Regulatory design: Transparency requirements (e.g., explanations mandated by DSA/AI Act) are a useful but partial policy lever. Regulators should pair disclosure with stronger privacy protections, accountability mechanisms, and technical safeguards to address concerns that explanations do not resolve.
• Welfare and market structure: Improved legitimacy via explanations can reduce frictions in consumer interactions with algorithms, potentially increasing efficiency, but persistent distrust in high-stakes domains could slow diffusion and entrench market winners who better manage privacy and governance.
• Cost–benefit assessment: Policymakers evaluating explanation mandates should account for modest behavioral returns and weigh them against compliance costs for firms; complementary interventions (privacy guarantees, audits, redress) will likely increase net social benefits.
• Future research priorities: linking survey-measured changes in willingness to delegate to real-world adoption behavior, quantifying heterogeneity by stakes of decisions and by demographic groups, and studying combinations of explanations with stronger privacy guarantees or accountability measures to maximize uptake.