Event Details

Title: Public trust in AI: no role for explanations

Abstract: There is a prevailing narrative that AI offers an enormous opportunity for positive social impact across any number of problem areas, improving our daily lives while driving economic growth, but that distrust of AI might prevent us from fully realising this potential. This is what underlies the significant focus on developing tools and strategies for promoting public trust in AI. Typical approaches assume the need for greater transparency so that individuals can see that the AI is indeed trustworthy, with many seeking to develop accessible public-facing explanations of how AIs work. In this talk I will argue that there is no direct role for explanations in promoting public trust, that explanations would indeed be detrimental to trust. More broadly, I will argue that Explainable AI approaches have imported a trust model that is entirely unsuitable for the context of public trust in AI, and will provide an overview of a more appropriate model and its implications for fostering public trust in AI. This talk offers a theoretical scaffolding for understanding the importance of AI regulation, and AI documentation that enables this regulation, as the only path to promoting public trust in AI. It also attempts to unravel the faulty and dangerous narrative that trust in AI is important only for instrumental reasons, when in fact trust and distrust are key signals as to whether AI lives up to our moral standards and contributes to a society we want to live in.

Teams link