Abstract for: A Systems Thinking Approach to Algorithmic Fairness

The current challenges in AI around algorithmic bias, fairness, and discrimination require taking a systems thinking approach to the problem. In this presentation, we will discuss how to apply systems thinking to bridge together the fields of machine learning, causal inference, and system dynamics. This will allow us to visualize and model how bias arises from the data generating process, which may result in harmful discrimination and the need for interventions into a machine learning model to make it fair. We will show how to use system dynamics to model bias from the data generating process and capture counterintuitive behavior using system archetypes. Finally, we will provide a system map of fairness to show how the different fields in the social sciences interact with the technical aspects of the problem.