Partial Identification Approach to Counterfactual Fairness Assessment
This addresses fairness assessment in AI decision-making for domains like criminal justice, providing a method to handle non-identifiability, but it is incremental as it builds on existing counterfactual fairness measures.
The paper tackles the challenge of evaluating counterfactual fairness measures when they are not identifiable from data by using partial identification to derive bounds on these measures from observational data, demonstrating on the COMPAS dataset that race changes have a positive spurious effect and age changes have a negative direct causal effect on recidivism risk scores.
The wide adoption of AI decision-making systems in critical domains such as criminal justice, loan approval, and hiring processes has heightened concerns about algorithmic fairness. As we often only have access to the output of algorithms without insights into their internal mechanisms, it was natural to examine how decisions would alter when auxiliary sensitive attributes (such as race) change. This led the research community to come up with counterfactual fairness measures, but how to evaluate the measure from available data remains a challenging task. In many practical applications, the target counterfactual measure is not identifiable, i.e., it cannot be uniquely determined from the combination of quantitative data and qualitative knowledge. This paper addresses this challenge using partial identification, which derives informative bounds over counterfactual fairness measures from observational data. We introduce a Bayesian approach to bound unknown counterfactual fairness measures with high confidence. We demonstrate our algorithm on the COMPAS dataset, examining fairness in recidivism risk scores with respect to race, age, and sex. Our results reveal a positive (spurious) effect on the COMPAS score when changing race to African-American (from all others) and a negative (direct causal) effect when transitioning from young to old age.