Program fairness and transparency: A formal methods perspective
Recent years have seen a huge shift in the kind of programs that most programmers write. Programs are increasingly data driven instead of being algorithm driven. They use various forms of machine learning techniques to build models from data, for the purpose of “decision-making”. In this talk, I will present the first verification tool for automatically certifying that a decision-making program satisfies a given fairness property, an important property given the various forms of bias is typically present in most datasets. In the second part of the talk, I will show an instance of how to make decision-making programs transparent via a program debugging formulation that is based on Pearl’s theory of causation.