Foundation Models at Work: Fine-Tuning for Fairness in Algorithmic Hiring

TL;DR

This paper introduces AutoRefine, a reinforcement learning-based method for fine-tuning foundation models to improve fairness in algorithmic hiring by reducing linguistic biases in job descriptions.

Contribution

The paper presents AutoRefine, a novel automated fine-tuning approach that leverages direct performance feedback to mitigate biases in language models for fair hiring practices.

Findings

AutoRefine effectively reduces linguistic biases in job descriptions.

The method improves diversity in candidate matching.

Experiments show positive results on real-world hiring data.

Abstract

Foundation models require fine-tuning to ensure their generative outputs align with intended results for specific tasks. Automating this fine-tuning process is challenging, as it typically needs human feedback that can be expensive to acquire. We present AutoRefine, a method that leverages reinforcement learning for targeted fine-tuning, utilizing direct feedback from measurable performance improvements in specific downstream tasks. We demonstrate the method for a problem arising in algorithmic hiring platforms where linguistic biases influence a recommendation system. In this setting, a generative model seeks to rewrite given job specifications to receive more diverse candidate matches from a recommendation engine which matches jobs to candidates. Our model detects and regulates biases in job descriptions to meet diversity and fairness criteria. The experiments on a public hiring dataset and a real-world hiring platform showcase how large language models can assist in identifying and mitigation biases in the real world.