Assessing Cognitive Biases in LLMs for Judicial Decision Support: Virtuous Victim and Halo Effects

Research examining five major LLMs found they exhibit human-like cognitive biases when evaluating judicial scenarios, showing stronger virtuous victim effects but reduced credential-based halo effects compared to humans. The study suggests LLMs may offer modest improvements over human decision-making in judicial contexts, though variability across models limits current practical application.

• →Five major LLMs including ChatGPT, Claude, and Gemini were tested for cognitive biases in judicial decision-making scenarios.
• →LLMs showed larger virtuous victim effects but significantly reduced credential-based halo effects compared to human benchmarks.
• →Models demonstrated no statistically significant penalty for adjacent consent situations.
• →Variability across different models and outputs currently restricts their use in judicial applications.
• →Overall results suggest modest improvements over human decision-making despite existing limitations.