Who to Trust, How and Why: Untangling AI Ethics Principles, Trustworthiness and Trust

Trust in AI is more complex than just system reliability: The paper emphasizes that trusting an AI system involves not only confidence in how the system performs, but also trust in the people who developed it—a distinction often overlooked in discussions about AI trustworthiness.

Popular AI ethics principles may not work as assumed: Features like explainability and transparency are frequently promoted as ways to build user trust, but the paper finds that empirical evidence showing these features actually increase how trustworthy users perceive AI systems is surprisingly limited and unclear.

People matter as much as technology: AI systems should be understood as "socio-technical systems" where the designers, developers, deployers, and users are equally important to the actual technology in determining whether an AI system is truly trustworthy—not just the algorithms themselves.

"Trustworthy AI" risks becoming meaningless: Without clearer definitions and distinctions between concepts like "trust," "trustworthiness," and specific ethical principles, the term "trustworthy AI" risks becoming a catch-all label for any desirable AI feature, making it harder to actually evaluate and improve AI systems.

We need more real-world evidence: The paper calls for gathering empirical data on what actually influences people's trusting behaviors with AI, rather than relying on assumptions about what should build trust.