Tell Me What You Know About Sexism: Expert-LLM Interaction Strategies and Co-Created Definitions for Zero-Shot Sexism Detection

2025-04-21

AI Summary

Current zero-shot gender bias detection suffers from insufficient collaboration between domain experts and large language models (LLMs). Method: We propose “co-authored definition,” a human-AI collaboration paradigm featuring a four-stage bidirectional workflow—knowledge elicitation, expert drafting, LLM generation, and joint refinement—to systematically integrate expertise from gender bias specialists with GPT-3.5 and GPT-4o. Contribution/Results: Evaluated on 2,500 multi-source texts across 67,500 zero-shot classification trials, our approach significantly improves detection accuracy for novice experts—sometimes surpassing both expert-only and LLM-only baselines. Co-authored definitions are longer and more semantically complex, reflecting deeper knowledge integration. This work pioneers a reproducible methodology for collaborative concept definition between domain experts and LLMs, offering a scalable, low-resource framework for bias detection in under-resourced settings.

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Humans and AI: Human-AI Collaboration / Human-AI Teaming
• Machine Learning: Ethics, Bias, and Fairness

Application Category

• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Social Networks and Social Media: Fairness and bias in social network and social media analysis

Abstract

This paper investigates hybrid intelligence and collaboration between researchers of sexism and Large Language Models (LLMs), with a four-component pipeline. First, nine sexism researchers answer questions about their knowledge of sexism and of LLMs. They then participate in two interactive experiments involving an LLM (GPT3.5). The first experiment has experts assessing the model's knowledge about sexism and suitability for use in research. The second experiment tasks them with creating three different definitions of sexism: an expert-written definition, an LLM-written one, and a co-created definition. Lastly, zero-shot classification experiments use the three definitions from each expert in a prompt template for sexism detection, evaluating GPT4o on 2.500 texts sampled from five sexism benchmarks. We then analyze the resulting 67.500 classification decisions. The LLM interactions lead to longer and more complex definitions of sexism. Expert-written definitions on average perform poorly compared to LLM-generated definitions. However, some experts do improve classification performance with their co-created definitions of sexism, also experts who are inexperienced in using LLMs.

Problem

• Investigates collaboration between sexism researchers and LLMs for zero-shot detection
• Evaluates performance of expert-written, LLM-generated, and co-created sexism definitions
• Analyzes 67,500 classifications to compare definition effectiveness in detection

Innovation

• Hybrid intelligence pipeline for sexism detection
• Expert-LLM co-created definitions improve classification
• Zero-shot classification with diverse definition prompts