Fair In-Context Learning via Latent Concept Variables

TL;DR

This paper proposes a method using latent concept variables to improve fairness in in-context learning with large language models on tabular data, reducing bias and discrimination.

Contribution

It introduces a novel latent concept variable approach for fair demonstration selection in in-context learning, enhancing fairness in high-stakes applications.

Findings

Improved fairness metrics on tabular datasets.

Effective reduction of bias through data augmentation.

Generalization of latent concepts across different LLMs.

Abstract

The emerging in-context learning (ICL) ability of large language models (LLMs) has prompted their use for predictive tasks in various domains with different data types, including tabular data, facilitated by serialization methods. However, with increasing applications in high-stakes domains, it has been shown that LLMs can inherit social bias and discrimination from their pre-training data. In this work, we investigate inherent bias in LLMs during in-context learning with tabular data. We focus on an optimal demonstration selection approach that utilizes latent concept variables for resource-efficient task adaptation. We design data augmentation strategies that reduce the correlation between predictive outcomes and sensitive variables, helping promote fairness during latent concept learning. We utilize the learned concept to select demonstrations and obtain fair predictions. The latent concept variables are learned using a smaller internal LLM and generalized to larger external LLMs. We empirically verify that the fair latent variable approach improves fairness results on tabular datasets compared to multiple heuristic demonstration selection methods.