---
id: "claim-bias-suspicion-increases-avoidance"
type: "claim"
source_timestamps: ["\\\"§ Sometimes", "people don’t want to know.\\\""]
tags: ["algorithmic-bias", "ethics"]
related: ["concept-moral-quandary-avoidance", "concept-willful-ignorance-in-ai"]
confidence: "high"
testable: true
speakers: ["Alex Chan"]
sources: ["adoption"]
sourceVaultSlug: "hbr-seg-adoption"
originDay: 9
articleStem: "hbr-edu-37-employees-not-questioning-ai"
sourceUrl: "https://hbr.org/2026/06/employees-arent-questioning-ai-advice-enough"
sourceTitle: "Employees Aren’t Questioning AI Advice Enough"
---
# Suspicion of algorithmic bias increases explanation avoidance

**Confidence:** high · **Testable:** yes · **Attributed to:** [[entity-alex-chan|Alex Chan]]

Contrary to the assumption that people want to root out bias, **warning users that an AI explanation might reveal racial or gender bias actually increases their avoidance of that explanation.** The study showed that avoidance rates **rose by more than 10 percentage points (to 23%)** when participants were told the explanation might indicate demographic influence. Users actively avoid information that creates moral discomfort.

This is the direct evidence for [[concept-moral-quandary-avoidance]] and a specific case of [[concept-willful-ignorance-in-ai]]. It presumes familiarity with [[prereq-algorithmic-bias]].

**Enrichment note:** Supported directionally — Chan's article reports that when fairness auditing is made salient and explanations may involve race/gender, lender-aligned participants are more likely to skip explanations; Meyer notes we should not expect explanations to be consulted "if an explanation threatens the interests of the person receiving it." **The specific "to 23%" figure is not independently corroborated in public summaries and should be treated as provisional.** Note the boundary condition surfaced in the counter-perspectives: the avoidance effect is strongest among *financially aligned* participants; with neutral pay, salient bias can instead *increase* override (see [[concept-algorithmic-override]]).