2023/06/07 by Himanshu Thakur, Atishay Jain, Thakur, Himanshu +7 · 6 citations
Computer Science · Medicine · Psychology · Social Sciences · #Artificial Intelligence in Healthcare and Education #Artificial intelligence #Cognitive psychology #Computation and Language (cs.CL) #Computer science #Data science #Debiasing #Ethics and Social Impacts of AI #FOS: Computer and information sciences #Gender bias #Language model #Machine Learning (cs.LG) #Machine learning #Natural language processing #Political science #Psychological intervention #Psychology #Rendering (computer graphics) #Retraining #Social psychology #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2306.04597
published in arXiv (Cornell University) (Cornell University)
openalex publication_date 2023/06/07 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/06
Societal biases present in pre-trained large language models are a critical issue as these models have been shown to propagate biases in countless downstream applications, rendering them unfair towards specific groups of people. Since large-scale retraining of these models from scratch is both time and compute-expensive, a variety of approaches have been previously proposed that de-bias a pre-trained model. While the majority of current state-of-the-art debiasing methods focus on changes to the training regime, in this paper, we propose data intervention strategies as a powerful yet simple technique to reduce gender bias in pre-trained models. Specifically, we empirically show that by fine-tuning a pre-trained model on only 10 de-biased (intervened) training examples, the tendency to favor any gender is significantly reduced. Since our proposed method only needs a few training examples, our few-shot debiasing approach is highly feasible and practical. Through extensive experimentation, we show that our debiasing technique performs better than competitive state-of-the-art baselines with minimal loss in language modeling ability.