Partial debiasing can make bias problem worse by hiding it

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

Categorization

• QUESTION: Does superficial debiasing provide false assurance that worsens downstream harm, and how should we measure implicit associations that remain?
• SEED: Strong hook for an essay on how measurable fixes can be counterproductive when they mask what remains.
• corroborated: 2 independent passages support this claim — Brian Christian, The Alignment Problem, 210
• XREF: connects to: Prior biased policing leaks into race-neutral variables like conviction history — Brian Christian, The Alignment Problem, 296; Removing protected attributes fails to remove bias from models — Brian Christian, The Alignment Problem, 294; Common language carries more gender bias than objective metrics show — Brian Christian, The Alignment Problem, 221 (+5 more)

Broader Connections

• Removing protected attributes fails to remove bias from models #corroborates (0.90) — Brian Christian, The Alignment Problem, 294
• Binary schematization’s completeness claim masks the imposition behind any dichotomy #expands (0.82)
• Prior biased policing leaks into race-neutral variables like conviction history #expands (0.70) — Brian Christian, The Alignment Problem, 296