Why Fair Labels Can Yield Unfair Predictions: Graphical Conditions for Introduced Unfairness
This addresses fairness issues in machine learning for affected groups by identifying conditions that prevent introduced unfairness, though it is incremental as it builds on existing causal and graphical fairness frameworks.
The paper tackles the problem of machine learning systems introducing or amplifying discriminatory effects beyond those in the training data, termed introduced unfairness, and establishes graphical conditions under which this occurs, showing that adding the sensitive attribute as a feature can remove the incentive for such unfairness under well-behaved loss functions.
In addition to reproducing discriminatory relationships in the training data, machine learning systems can also introduce or amplify discriminatory effects. We refer to this as introduced unfairness, and investigate the conditions under which it may arise. To this end, we propose introduced total variation as a measure of introduced unfairness, and establish graphical conditions under which it may be incentivised to occur. These criteria imply that adding the sensitive attribute as a feature removes the incentive for introduced variation under well-behaved loss functions. Additionally, taking a causal perspective, introduced path-specific effects shed light on the issue of when specific paths should be considered fair.