Trust in AI: Progress, Challenges, and Future Directions

Trust acts as a critical adoption regulator: The paper identifies trust/distrust in AI as a key factor that can significantly control how quickly AI systems are adopted by users—trust increases adoption rates while distrust reduces them, making it essential for responsible AI deployment.

AI trustworthiness requires both technical and non-technical measures: The research proposes a taxonomy distinguishing between technical trustworthiness metrics (safety, accuracy, robustness) and non-technical axiological ones (ethical, legal, and mixed considerations), indicating that responsible AI cannot rely on performance metrics alone.

AI systems pose unique trust challenges compared to other technologies: Unlike traditional tools, AI systems are increasingly acting as substitutive agents making decisions on our behalf or as "manipulative minds" influencing human thought and decision-making, which creates distinct trust concerns that require specific attention.

Specific trust-breakers must be addressed: The paper identifies autonomy and dignity threats as major factors that erode trust in AI systems, suggesting that responsible AI development must actively protect user autonomy and human dignity to maintain public confidence.

Human-machine interaction context matters for trust: Trust in AI varies significantly depending on the type of interaction and application domain, implying that one-size-fits-all approaches to trustworthy AI are insufficient and solutions must be tailored to specific use cases.