Proxy Non-Discrimination in Data-Driven Systems
This addresses fairness and bias issues in AI systems, particularly for historically disparaged groups, with a formal approach that is incremental but practical.
The paper tackles the problem of machine learning systems inheriting subtle biases against protected classes through proxy discrimination, formalizing it as the presence of protected class correlates with causal influence on outputs, and demonstrates validation and repair methods on social datasets.
Machine learnt systems inherit biases against protected classes, historically disparaged groups, from training data. Usually, these biases are not explicit, they rely on subtle correlations discovered by training algorithms, and are therefore difficult to detect. We formalize proxy discrimination in data-driven systems, a class of properties indicative of bias, as the presence of protected class correlates that have causal influence on the system's output. We evaluate an implementation on a corpus of social datasets, demonstrating how to validate systems against these properties and to repair violations where they occur.