Aligned but Blind: Alignment Increases Implicit Bias by Reducing Awareness of Race

TL;DR

Aligning language models to human values can unintentionally increase implicit racial bias by reducing the model's awareness of racial concepts, which can be mitigated by targeted early-layer interventions.

Contribution

This paper reveals that alignment amplifies implicit bias by diminishing racial concept awareness and proposes a novel mitigation strategy focusing on early-layer representation of race.

Findings

Alignment increases implicit racial bias in language models.

Models overlook racial concepts in ambiguous contexts after alignment.

Early-layer interventions reduce implicit bias effectively.

Abstract

Although value-aligned language models (LMs) appear unbiased in explicit bias evaluations, they often exhibit stereotypes in implicit word association tasks, raising concerns about their fair usage. We investigate the mechanisms behind this discrepancy and find that alignment surprisingly amplifies implicit bias in model outputs. Specifically, we show that aligned LMs, unlike their unaligned counterparts, overlook racial concepts in early internal representations when the context is ambiguous. Not representing race likely fails to activate safety guardrails, leading to unintended biases. Inspired by this insight, we propose a new bias mitigation strategy that works by incentivizing the representation of racial concepts in the early model layers. In contrast to conventional mitigation methods of machine unlearning, our interventions find that steering the model to be more aware of racial concepts effectively mitigates implicit bias. Similar to race blindness in humans, ignoring racial nuances can inadvertently perpetuate subtle biases in LMs.