This revolutionary paper dives into the dynamics of Large Language Models (LLMs) armed with distinct personality traits, engaging in both single and multi-issue negotiations. The authors, Sean Noh and Ho-Chun Herbert Chang, utilize game-theoretical frameworks to analyze how traits like openness, conscientiousness, and neuroticism influence negotiation strategies, presenting their findings with robust statistical tools such as gradient-boosted regression and Shapley explainers. Major highlights of the study include:
These granular insights hint at the dual potential of LLMs; while they naturally lean towards equitable interactions, they can be manipulated to gain unsporting advantages against compliant opponents. This study not only contributes to understanding LLMs’ behavioral economics in AI negotiations but also prompts further exploration into ethical guardrails for AI personalities in automated negotiations.
Read the full paper here for more detailed results and analysis.