Calls for making AI safe, trustworthy and fair are made by scholars, legal institutions, NGOs, and customer protection services alike. However, how to best satisfy these desiderata remains the subject of academic, political, and legal debates. Over the past decade or so, research on explainable AI (XAI) has taken center stage in these debates. While there has been a veritable XAI-hype, and XAI clearly can contribute to desiderata satisfaction in some cases, it must be acknowledged that XAI is not a silver bullet. There are a number of reasons why XAI may fail to successfully lead to desiderata satisfaction, and there are a number of issues XAI cannot successfully address. This talk offers a systematic analysis of XAI shortcomings from an interdisciplinary perspective incorporating contributions from philosophy, law, computer science and psychology. It sketches possible ways forward by proposing strategies that might be needed in addition to XAI research. Eventually, a clever combination of different methods and strategies will hopefully yield satisfaction of important societal desiderata.
Lena Kästner (Thu,) studied this question.