XAI #7-Work­shop: Eth­ics and Norm­ativ­ity of Ex­plain­able AI

XAI #7-Workshop of TRR318: “Constructing Explainability" | Paderborn University | May 15-17 2024

 

Organisers: Tobias Matzner, Suzana Alpsancar, Martina Philippi, Wessel Reijers.

 

Explanations are not per se helpful or wanted, they can also be a burden. They are not of absolute but of instrumental value. For the field of explainable AI, it is crucial to determine when and for whom which explanation will most likely be valuable and when not. This normativity of explanations can generally be grasped from an economic point of view (efficiency, comfort), from a social point of view (roles, constellations, contexts, stakeholder interests), and from an ethical-political point of view (obligations, duties, virtues, impacts of (not) explaining). The workshop serves to sort out these different normative expectations and implications and tries to link them to the task of value-oriented technology design. The urgency of this workshop is emphasised by the proliferation of AI systems and their use in decision making processes, from credit score assessments, via risk assessment models, to judicial rulings.

We expect a wide range of perspectives from philosophy of technology, media studies, STS, law, engineering, computer science interested in exploring the normativity of xAI.

Particularly, we are going to discuss the following issues and questions:

First, where are the limits of xAI? When is it helpful for users, needed institutionally or even ethically mandatory and when not? When does it draw the attention away from other ethical, e.g., transparency, concerns? When does it overburden users?

Second, we would like to explore the task of operationalizing ethical demands, such as explainability, and the context-dependency of the value and adequateness of explanations. In how far can existing ethical frameworks cover contextual factors, how would they need to be modified and where are the limits of designing for situated actions? How should we link xAI and conceptual engineering? How to not only bridge the gap between principles and practices but also between what is designable by technological features and what rests on contextual factors such as structural conditions, social roles, individual behavior, implicit and explicit norms governing certain practices.

Third, we like to address the interplay of ethical considerations and regulation, asking how xAI can inform ethical design methods and how it plays a role in upcoming regulations like the EU AI Act.  

 

Essential information: