A bias-removing model should predict counterfactual race-neutral outcomes, not likely ones.

Evidence

The approach we are suggesting here is distinct, as the goal is to remove suspect features, such as racial bias, from the model, thus bringing its predictions into closer accord with Constitutional mandates for racially equitable treatment of criminal defendants by state actors. Rather than predict likely outcomes, the objective for the model is to suggest how the case would resolve if the defendant were treated race-neutrally. For this reason, we call the model a “suggestive model.” — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law, 1219

Categorization

• DEFINE: Distinguishes the ‘suggestive model’ (predicting race-neutral counterfactual outcomes) from standard predictive models that include bias as a feature.
• corroborated: 2 independent passages support this claim — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law: A Definitive Look at…, 1219