'Challenges of Explainable AI' by David Danks (UC San Diego) 11 June 2025, 18:00 (Rome time) - Phinance Online Seminars

David Danks (University of California, San Diego)

Challenges of Explainable AI

by David Danks (UC San Diego)

11 June 2025, 18:00 - Rome time

Phinance Online Seminars are organized by Phinance, the Philosophy & Finance Network

Open to the public. No registration required

Zoom link: https://uniroma1.zoom.us/j/83417748694

18:00 David Danks (UC San Diego) Challenges of Explainable AI

18:45 Debate

organization & chair: Emiliano Ippoliti (Sapienza University of Rome)

As AI systems are becoming increasingly common in the finance sector, there is a growing need (including regulatory demands) for them to provide explanations of their model outputs. Explainable AI is also thought valuable because it can potentially help to reduce biases or increase trust. In this talk, I will first outline the reasons why people think that explainable AI might provide significant psychological, societal, and ethical benefits. However, I will then present a series of experiments that point towards a much more complicated cognitive picture than is standardly assumed in the field of explainable AI. I will argue that people are goal- and norm-directed in ways that fail to match most methods for deriving explanations for AI. We need a new, different type of approach to explainable, and justifiable, AI systems.

An open debate follows the meeting.