The Impact of AI Trustworthiness Labels on the Perception of AI Products

Most often, potential users aren't well informed about the trustworthiness of AI products when selecting them. They may therefore show misplaced (dis-)trust towards AI products. Here, participants were presented with (hypothetical) AI products (smart fridges, voice assistants) of (hypothetical) brands that were paired with a graphical, traffic light-like label conveying low to high AI trustworthiness or they were presented with AI products without such a label (baseline). Higher trustworthiness levels as indicated by the AI trustworthiness labels increased trust, acceptance, and the intention to use AI products, as well as the monetary value participants attributed to AI products, but did not affect the evaluation of an AI product's brand. These findings suggest that labels effectively communicated differences in AI trustworthiness. Interestingly, the evaluation of unlabeled AI products corresponded to AI products labeled with an intermediate trustworthiness level, highlighting biased assessment and the need to communicate AI trustworthiness to potential users.

• Human-centered computing → Human computer interaction (HCI); Empirical studies in HCI; Human computer interaction (HCI); HCI design and evaluation methods; User studies; • Social and professional topics → Computing / technology policy; Government technology policy; Governmental regulations Security and privacy; Human and societal aspects of security and privacy; Privacy protections.