Talk Michel Besserve

Abstract:

Our societies are increasingly relying on artificial intelligence to help humans perform complex tasks by harnessing vast amounts of data. However, designing more reliable and explainable machine learning algorithms is essential to unlocking their potential in many applications. These desiderata can be framed in terms of changes to the data generation process: we "understand" and trust a system when we know how it behaves in the face of plausible and meaningful changes in its environment. Causality provides a comprehensive framework for modelling these changes, through the concepts of interventions and counterfactuals. In this talk, I will show how we can build causal generative AI tools, capable of representing meaningful changes in complex artificial, physical and socioeconomic systems. I will further demonstrate the potential of these tools for promoting the reliability of AI methods and helping humans solve complex problems in the real world.