Activation Steering for Bias Mitigation: An Interpretable Approach to Safer LLMs

TL;DR

This paper introduces an interpretable activation steering method that detects and mitigates biases within large language models by manipulating internal activations, leading to safer and more accountable AI outputs.

Contribution

The work presents a novel end-to-end bias mitigation system using mechanistic interpretability to identify and actively steer away from biased content within LLMs.

Findings

Probes accurately detect bias in GPT-2 layers

Steering vectors effectively reduce stereotypical outputs

Bias mitigation is achieved in real-time during inference

Abstract

As large language models (LLMs) become more integrated into societal systems, the risk of them perpetuating and amplifying harmful biases becomes a critical safety concern. Traditional methods for mitigating bias often rely on data filtering or post-hoc output moderation, which treat the model as an opaque black box. In this work, we introduce a complete, end-to-end system that uses techniques from mechanistic interpretability to both identify and actively mitigate bias directly within a model's internal workings. Our method involves two primary stages. First, we train linear "probes" on the internal activations of a model to detect the latent representations of various biases (e.g., gender, race, age). Our experiments on \texttt{gpt2-large} demonstrate that these probes can identify biased content with near-perfect accuracy, revealing that bias representations become most salient in the model's later layers. Second, we leverage these findings to compute "steering vectors" by contrasting the model's activation patterns for biased and neutral statements. By adding these vectors during inference, we can actively steer the model's generative process away from producing harmful, stereotypical, or biased content in real-time. We demonstrate the efficacy of this activation steering technique, showing that it successfully alters biased completions toward more neutral alternatives. We present our work as a robust and reproducible system that offers a more direct and interpretable approach to building safer and more accountable LLMs.