Research Summary: When Does Defendant Statement Matter? A Study of Bias and Persuasion in LLM-Simulated Jurors

Research from arXiv explores the use of Large Language Models (LLMs) to simulate juror decision-making, specifically examining how defendant statements influence LLM-simulated jurors. The study investigates factors such as persuasion, ideological bias, and background-based affinity in common-law jury trials. It introduces 'JuryBench,' a benchmark comprising controversial U.S. criminal cases, to test varied defendant backgrounds and courtroom statements (emotional appeal, rebuttal) against diverse ideological juror profiles.

Why it matters

This research provides insights into the potential for artificial intelligence to model and predict human decision-making in high-stakes environments. Understanding how simulated jurors are influenced can inform future developments in legal technology, bias mitigation, and the strategic presentation of information in various decision-making contexts.

Key insights

• LLMs are being employed to simulate human decision-making, including complex scenarios like jury trials.
• The study focuses on the impact of defendant courtroom statements on LLM-simulated jurors.
• Key areas of investigation include persuasion, ideological bias, and background-based affinity.
• A new benchmark, 'JuryBench,' has been developed using controversial U.S. criminal cases.
• The research design involves varying defendant backgrounds and statement characteristics (emotional appeal, rebuttal).
• Simulated jurors are designed with diverse ideological profiles to assess spectrum-wide responses.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.09887