Personality Traits Shape LLM Bias in Decision-Making
How model personality influences cognitive bias and affects security applications
This research investigates how personality traits in Large Language Models affect the manifestation of cognitive biases in automated decision-making tasks.
Key Findings:
- Six prevalent cognitive biases were identified in LLMs across different decision contexts
- Personality traits significantly influence bias expression in AI decision-making
- Sunk cost and group attribution biases showed minimal impact in LLMs
- Certain mitigation strategies proved effective across model architectures
Security Implications: Understanding personality-driven biases is crucial for developing reliable AI systems for security applications where fair, unbiased decision-making is essential. This research provides a foundation for creating more trustworthy automated decision-making systems in sensitive contexts.