Exposing AI Bias by Crowdsourcing: Democratizing Critique of Large Language Models

Résumé fourni par la source

The widespread adoption of large language models (LLMs) and generative AI (GenAI) tools across diverse real-world applications has amplified the importance of addressing societal biases inherent within these technologies. While the Natural Language Processing (NLP) community has extensively studied LLM bias, research investigating how non-expert users perceive and interact with biases from these systems remains limited. As these technologies become increasingly prevalent, understanding this question is crucial to inform model developers in their efforts to mitigate bias. To address this gap, this paper presents findings from a university-level competition that challenged participants to design prompts specifically for eliciting biased outputs from GenAI tools. We conducted a quantitative and qualitative analysis of the submitted prompts and the resulting GenAI outputs. This analysis led to the identification of reproducible biases across eight distinct categories within GenAI systems. Furthermore, we identified and categorized the various strategies employed by participants to successfully induce these biased responses. Our findings provide unique insights into how non-expert users understand, engage with, and attempt to manipulate biases in GenAI tools. This research contributes to a deeper understanding of the user-side experience of AI bias and offers actionable knowledge for developers and policymakers working towards creating fairer and more equitable AI systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.