Causality and Responsibility in AI

Causality and responsibility are intertwined concepts that play an important role in the reasoning of human and artificial intelligent (AI) systems. These concepts have been extensively studied in the literature, resulting in a plethora of views and interpretations. Although the idea of causality and responsibility has been around for a while, recent developments in data-driven AI using machine learning techniques make the link between causality, correlation and responsibility urgent and challenging. Moreover, there still exist contradictory views on responsibility attribution when a group of individual actors have contributed towards causing an effect or how to analyse causal relations that develop over a longer period. Such issues are especially important when discussing causation in law. 

This workshop brings together researchers from AI, machine learning, logic, mathematics and law to discuss about the conceptual foundation of causality and responsibility and how they should be formalized with the support of mathematical/computational tools. The following three aims can be identified:

1. A cartography of the landscape of research on causality andresponsibility from different perspective (specifically, philosophical, legal, ethical, and mathematical). Comparing and aligning various proposals from different disciplines.

1. Causality and responsibility of AI systems: a survey of formal andcomputational models for learning and reasoning about causalityand responsibility. Compare various AI challenges and models from the Artificial Intelligence perspective.

1. Identifying practical use cases and application areas in legal assessment of responsibility for the different models of causality.

The following is a non-exhaustive list of research questions that could be addressed during the workshop: Which concepts andmodels of cause, responsibility and causal explanation are the most appropriate for moral and legal responsibility? What are their distinctive formal properties? What are their computational complexities? Which are the advantages and limitations of each of the existing mathematical/computational tool for modeling causality and responsibility in comparison? Is it possible to define a common base, to integrate the different approaches to causality, and to find formal translations from one to another? On this basis, is it possible to provide workable concepts of causality, to be used inpractical contexts, such as in criminal and civil law cases?