The webinar on Trustworthy AI hosted by the University of Copenhagen is part of the webinar series in the context of the EU project DisAI. The webinar was led by Prof. Anders Søgaard (University of Copenhagen).

The webinar gave a high-level overview of the fundamental problems in trustworthy AI. Trustworthiness is said to track transparency, fairness, privacy, and accountability, with transparency often receiving more attention. Transparency comes in two distinct flavors, inference and training transparency. 

The talk first focused on the classical problem of inference transparency, and why it is a hard problem. It also looks at how this problem affects the epistemic value of applications of AI to the sciences – as well as at possible ways out. 

Finally, it addressed some common fallacies when addressing the trustworthiness of AI systems, in general.

You can watch the webinar recording here: