How might people assess whether an AI system is trustworthy?

Details

Designing trustworthy AI-based systems and enabling outside parties to assess their trustworthiness accurately are important goals. Accurate assessment allows people to base their trust on appropriate expectations and reasonably rely on or reject a system’s outputs. Yet how people assess a system’s actual trustworthiness and arrive at their perceived trustworthiness remains underexplored.

In this conceptual paper, the authors distinguish actual trustworthiness, perceived trustworthiness, trust propensity, trust, and trusting behavior. Drawing on psychological models of how people assess others’ characteristics, they present the two-level Trustworthiness Assessment Model (TrAM).

At the micro level, the authors propose that people assess a system’s trustworthiness using cues associated with it. Assessment accuracy depends on the relevance and availability of cues on the system’s side and on their detection and use on the human’s side. At the macro level, they propose that these individual assessments propagate among trustors: one stakeholder’s assessment of a system affects other stakeholders’ assessments of the same system.

The authors say TrAM advances existing trust models and illuminates factors influencing trustworthiness assessments and their accuracy. They also describe contributions to theoretical clarity in trust research, implications for measuring trust-related variables, and practical implications for system design, stakeholder training, and regulation related to trustworthiness assessments.