Debiasing by marginalization uses all predictors but neutralizes protected variables at prediction time.

Evidence

In fact, such only partial debiasing may actually make the problem worse, they argue, in the sense that it leaves the majority of these stereotypical associations intact while removing the ones that are the most visible and easiest to measure. — Brian Christian, The Alignment Problem, 210 (#complicates, via linked claim)

Categorization

• XREF: connects to: Race-blind prediction models can encode bias; active debiasing improves fairness. — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law: A Definitive Look at…, 1185; Partial debiasing can make bias problem worse by hiding it — Brian Christian, The Alignment Problem, 210; Debiasing algorithms risk exacerbating bias without a careful implementation framework — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law: A Definitive Look at…, 1223 (+2 more)

Broader Connections

• Partial debiasing can make bias problem worse by hiding it #complicates (0.72) — Brian Christian, The Alignment Problem, 210