Healthy Distrust in AI systems

Trustworthy AI shouldn't automatically mean maximizing trust: The paper challenges the prevailing assumption in AI research that designing systems to inspire human trust is always beneficial. Instead, the authors argue that in contexts where AI systems may conflict with users' interests, skepticism can be rational and necessary.

"Healthy distrust" is justified caution, not blanket rejection: The paper introduces "healthy distrust" as a deliberate, careful skepticism toward certain AI practices—distinct from paranoia or blanket rejection of AI. This stance becomes particularly important when social and institutional contexts suggest an AI system might be used against a person's interests.

Trust built on distrust is more meaningful: According to the paper's framework, genuine trust in AI systems can only develop after people have exercised appropriate skepticism and verified that systems respect their interests. Skipping the distrust phase may result in false confidence rather than authentic trust.

Respecting human autonomy requires enabling critical evaluation: The paper positions healthy distrust as essential to AI usage that truly respects human autonomy. This suggests that responsible AI deployment should facilitate—rather than discourage—users' ability to question and scrutinize how systems affect them.

Current AI ethics frameworks have a gap: By drawing on research from computer science, sociology, history, psychology, and philosophy, the paper identifies that existing notions of trust and distrust don't adequately address situations where skepticism toward AI systems is both justified and protective of human interests.