Explanations and justifications in legal XAI

Abstract:
In this talk I will discuss my recent research with Rosa Ratsma on legal explainable (XAI) in a wider legal and AI context. I will first give an overview of both AI research on and legal requirements for explanations of algorithmic decision making. I will then outline our case-based method for explaining predictions of supervised-machine-learning applications, which draws on AI & law research on argumentation with cases. A case-based approach is natural since the training data of supervised machine-learning applications can be seen as cases. I will explain that our method strictly speaking does not *explain* but *justify* predictions, and I will discuss the place of justification methods in (legal) XAI. Finally I will put my discussions in the context of legal requirements for explanations, such as those of the GDPR.

Reference:
H. Prakken & R. Ratsma, A top-level model of case-based argumentation for explanation: formalisation and experiments. Argument and Computation 13 (2022): 159-194.