Presenter: Os Keyes

Abstract:

Algorithmic audits - in which "key stakeholders" work through the design of an artificial intelligence system to try to identify and then prevent injustices - are commonly discussed as a potent social and regulatory response to technological harms. But what do they look like - and feel like - in practice? In this talk, I will present one effort to answer that question: a duoethnography of auditing a machine learning dataset. Drawing on science and technology studies and feminist philosophy, I will discuss the limits and costs of algorithmic audits: not only the pragmatic difficulty of conducting them in infrastructure that is often distributed over multiple regulatory and organisational environments, but the affective and personal consequences of audit work, and who (specifically) bears those costs.