The value of vulnerability for trustworthy AI

Abstract

The concept of trustworthy AI plays a significant role in many policy recommendations and ethics guidelines, as it is implied to be a good compromise between the desire for technological innovation and the duty to respect citizens’ rights. However, the concept has often been criticized for its unclarity, its dubious ethical value, and has even been seen as a symptom of ethics washing. I aim to repurpose the value of trustworthy AI for this kind of technology, its development, deployment, and regulation by clarifying the relationship with vulnerability within social arrangements. I argue that vulnerability is one of the key reasons why we enter social arrangements to begin with, and that trust not only works as a facilitator of those social arrangements but requires the recognition and willingness to address vulnerability. Rethinking the trustworthiness of AI and its surrounding practices as a matter of aptly recognizing and addressing vulnerability leads to reconsidering not just the conceptual grounding of trustworthy AI, but also the responsibilities of sovereign actors and the priorities of AI governance. I conclude by exploring the practical implications of such a renewed conception of Trustworthy AI.