Expose Uncertainty, Instill Distrust, Avoid Explanations: Towards Ethical Guidelines for AI

Transparency about uncertainty is essential: Rather than presenting AI predictions as definitive answers, systems should explicitly communicate their confidence levels and limitations. This helps users understand when they should be cautious about relying on AI recommendations, particularly in high-stakes decisions.

Healthy skepticism toward AI is protective: The paper argues that fostering appropriate distrust in AI systems—rather than blind confidence—better protects users from potential harms. This contrasts with industry trends that emphasize user trust and adoption.

Detailed explanations can be counterproductive: Contrary to popular calls for "explainable AI," the author suggests that providing detailed explanations of how AI systems work may give users false confidence in their reliability or create an illusion of understanding that isn't warranted, potentially doing more harm than good.

Current AI presentation practices may mislead users: The paper critiques how AI results are typically presented to the public, suggesting that standard practices obscure the intrinsic limitations and problems of algorithms rather than exposing them.

Ethical AI requires rethinking user communication: Rather than focusing solely on making AI more interpretable or trustworthy, the author proposes that ethical AI deployment should prioritize honest communication about what AI cannot reliably do.