In Seal We Trust? Investigating the Effect of Certifications on Perceived Trustworthiness of AI Systems

Abstract

Trust certification through so-called trust seals is a common strategy to help users ascertain the trustworthiness of a system. In this study, we examined trust seals for AI systems from two perspectives: (1) In a pre-registered online study with N = 453 participants, we asked whether trust seals can increase user trust in AI systems, and (2) qualitatively, we investigated what participants expect from such AI seals of trust. Our results indicate mixed support for the use of AI seals. While trust seals generally did not affect the participants’ trust, their trust in the AI system increased if they trusted the seal-issuing institution. Moreover, although participants understood verification seals the least, they desired verifications of the AI system the most.