Foreword∗ . Human agency 0 oversight 0 . Technical robustness 0 safety 0 . Privacy 0 data governance 0 . Transparency 0 . Diversity 0 non-discrimination 0 fairness 0 . Societal 0 environmental wellbeing 0 . Accountability 0 Alessandro Sa otti , Luciano Sera ni, Paul Lukowicz Workshop chairs

Many of today's most popular AI methods, however, fail to meet these guidelines: making them compliant is a scienti c endeavor that is as crucial as it is challenging and stimulating. As an example, systems based on deep learning point often provide impressive results, but their ability to explain these results to the user is limited, thus challenging requirements 4 and 7; we lack general ways to formally verify their correctness and assess their boundary conditions, thus challenging requirement 2; and we don't yet have methods to allow humans to collaboratively in uence or question their decisions, thus challenging requirement 1. Similar criticalities are present in many other popular AI methods.