As part of sub-project ‘Ethics and Normativity of Explainable AI’, within the SFB/TRR 318 ‘Constructing Explainability’, the first funding period saw a re-examination of the widely held assumption that explanations for AI decisions are, in principle, desirable. The project team, comprising researchers from the fields of Philosophy and Media Studies, demonstrated instead that there are a variety of normative reasons for explainable AI (XAI) and developed a taxonomy of typical purposes: functional, economic, epistemic and ethical-political purposes. It also became clear that the relevance of explanations depends heavily on the specific context of use and that the XAI techniques developed to date do not do justice to this diversity of purposes and contexts.
When the context-dependence of XAI is taken into account, the focus shifts from the individual explanatory situation (micro-level) to the discursive practices within organisational contexts (meso-level). AI is no longer viewed in isolation as human-machine interaction, but rather as a socio-technical system embedded in specific institutional settings – such as hospitals, educational institutions or the police force.
In the second phase of the project, the relevance of explanations will be examined using selected scenarios from the fields of medicine, education and policing. To this end, the Value-Sensitive Design (VSD) method – which was already critically refined in Phase 1 – will be expanded to include the dimension of organisational contexts. The aim is to develop a practice-oriented approach to the ethics of XAI that incorporates an organisational level. This approach is intended to demonstrate how organisational frameworks influence epistemic and civic virtues, and what normative implications arise from the introduction of explainable AI in such contexts.
