One, none, one hundred thousand AIs

Abstract

Like many innovative technologies, AI possesses a transformational power: its implementation in society is not a neutral additive process, but it may alter in significant ways various social and cultural dynamics. Socio-ethical concerns led to the demand of designing AI devices that can be ‘trusted’. The recent publication of Machines we trust provides novel opportunities to discuss some socio-ethical issues arising from human-AI interactions. After defining the concepts of trust, trustworthiness, and reliability, and explaining in which sense it is possible to talk about ‘trustworthy AI’, I focus on two chapters of the volume that consider some concrete applications of AI. I conclude by suggesting that, instead of considering the different contributions to the volume in isolation with respect to one another, it may be illuminating to compare and contrast them. Such a way of reading the book leads us to question whether it is still possible to talk about trustworthy AI ‘in general’ or whether the discussion about the socio-ethical issues posed by AI should proceed in a piece-meal case-by-casefashion.