Understanding the Self-Reflection Mechanisms of LLMs through Biased Attitude Associations
For AI safety researchers, it reveals that self-reflection in LLMs can entrench rather than reduce certain biases, challenging the assumption that self-reflection is universally beneficial.
The paper proposes ReBias-Lens to probe how self-reflection in LLMs reconfigures biased attitude associations, finding that while overall bias is mitigated, category-specific biases are often amplified.
While the emergent self-reflection capabilities of Large Language Models (LLMs) offer a promising paradigm for autonomous bias mitigation, their internal mechanics remain unclear, raising concerns regarding potential bias entrenchment. Under the premise that social bias is intrinsically encoded as valence inclinations, where the exacerbation of bias scales with sharper valence fluctuations across social groups, this paper proposes ReBias-Lens, a probing framework designed to interpret how self-reflection reconfigures these biased attitude associations through the lens of valence projection within intersectional contexts. Central to ReBias-Lens is the metric of Valence Fluctuation (VF) comprising two variants: Global-VF, which captures macroscopic valence encoding trends, and Local-VF, which scrutinizes microscopic distinctiveness across specific social categories. Deploying ReBias-Lens to evaluate four LLMs across twelve social categories reveals that overall valence fluctuations undergo a distinct layer-wise smoothing, characterized by a significant hierarchical representation divergence as the layers deepen, which ultimately manifests as a widespread mitigation of bias at the behavioral level. In stark contrast to this macro-level reduction, this reflection mechanism is not universally corrective, instead exhibiting a stubborn, category-specific selectivity that regularly locks in and perversely amplifies localized biases. Warning: this paper contains examples with biased content.