A trustee’s evaluation, or perception, of whether, when, why, or to what degree someone or something should or should not be trusted.
Related terms: Trust, Interpretability, Explainability, Ethical AI, Responsible AI, Robust AI
The dimensions of trustworthiness are dependent on the values and needs of the trustor (Schneiderman, Human-Centered AI, 2022). A model deemed trustworthy by one user does not imply that it will be trustworthy to another user with their own set of values and needs. The common dimensions associated with evaluation of trustworthiness include: model performance and verification, trustor’s experience with model, source of information, and usability of model and information. Dimensions specific to ML are basic performance, reliability, human interaction, and aligned purpose (Varshney 2022). Note, the provided dimensions are meant to be representative examples but not exhaustive of all possible dimensions.