ANUBIS: For a Future where AI mirrors the best of us, free from the shadows of bias.

Anonymous

Abstract: Bias identification and mitigation is an important research problem with far-reaching societal impact. Though there exist datasets for bias mitigation, they offer superficial debiased gold-standard. In the scope of the paper we present a high-quality dataset (ANUBIS) for evaluation of debiasing across bias types in conjunction with LLMs and human annotators. In addition, we leverage advanced Large Language Models (LLMs) for automatic and effective bias detection and mitigation.

Paper Type: long

Research Area: Ethics, Bias, and Fairness

Contribution Types: NLP engineering experiment, Data resources

Languages Studied: English