ResearchBy merging causality with machine learning, our goal is to achieve fundamental breakthroughs in developing AI systems that are aligned to society's values.
|
Working Papers
Fairness Theory
We apply a sociotechnical approach to the fairness problem by directly mapping a principal cause of algorithmic bias (the structure and agency debate in sociology) to Judea Pearl's two fundamental laws of causal inference (counterfactuals/interventions and conditional independence).
|
Fairness Dynamics
We extend Judea Pearl's causal hierarchy with Jay Forrester's work in system dynamics. This allows us to model the complex emergent behavior between causality and machine learning to help explain why disparities across protected groups may develop and reinforce themselves over time.
|
Causal Data Science
We describe how to use Judea Pearl's tools of causal inference to solve practical problems in data science. Our patent-pending technology can help remove biases from data, prevent data from acting as an illegal proxy, and make interventions into a model to meet legal discrimination requirements.
|