No LLM is Free From Bias: A Comprehensive Study of Bias Evaluation in Large Language models

2025-03-15

AI Summary

This work addresses multidimensional social biases in large language models (LLMs) stemming from training data. We introduce the first unified bias evaluation framework covering diverse sensitive attributes—including physical characteristics and socioeconomic factors. To enable systematic assessment, we propose five generalizable prompting strategies for automated, cross-bias-type detection; design implicit bias metrics—Stereotype Score and Bias Amplification Ratio—and establish a multi-benchmark comparative evaluation framework. Empirical analysis across state-of-the-art models reveals that all exhibit statistically significant bias in at least one dimension, with LLaMA3.1-8B showing the lowest overall bias severity. Crucially, we identify data contamination and instruction-alignment mismatch as primary root causes. Our study delivers a rigorous, reproducible methodology and an open empirical benchmark for LLM bias evaluation and mitigation.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Computer Vision: Bias, Fairness & Privacy

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Social Networks and Social Media: Fairness and bias in social network and social media analysis

Abstract

Advancements in Large Language Models (LLMs) have increased the performance of different natural language understanding as well as generation tasks. Although LLMs have breached the state-of-the-art performance in various tasks, they often reflect different forms of bias present in the training data. In the light of this perceived limitation, we provide a unified evaluation of benchmarks using a set of representative LLMs that cover different forms of biases starting from physical characteristics to socio-economic categories. Moreover, we propose five prompting approaches to carry out the bias detection task across different aspects of bias. Further, we formulate three research questions to gain valuable insight in detecting biases in LLMs using different approaches and evaluation metrics across benchmarks. The results indicate that each of the selected LLMs suffer from one or the other form of bias with the LLaMA3.1-8B model being the least biased. Finally, we conclude the paper with the identification of key challenges and possible future directions.

Problem

• Evaluate biases in Large Language Models (LLMs).
• Propose methods for detecting various forms of bias.
• Identify challenges and future directions in bias evaluation.

Innovation

• Unified bias evaluation using representative LLMs
• Five prompting approaches for bias detection
• Research questions to detect biases in LLMs