Abstract

In this talk I will (briefly and informally) introduce the theory of actual causality as defined by Halpern and Pearl and its quantification. This theory turns out to be extremely useful for explaining outputs of computerised systems due to a good match between the results it produces and our intuition. 

We will then look in more detail at the application of actual causality to explainability of AI systems (XAI). I will present an overview of our tool ReX (Responsibility-based EXplanations) and discuss its variants and directions for future work. Time permitting, I will touch upon the problem of explanations of absence and how they can be addressed in the theory of actual causality and in our tool.  

The talk is based on a number of papers, and, while not strictly limited to my own research, the topics are quite broad, so I will be mostly talking about my work with different co-authors. The talk is reasonably self-contained and non-technical (as in “there will be no formulae on the slides”).