How Personas Can Influence Agents to Play Split or Steal

人格设定如何影响智能体在“平分或独吞”博弈中的决策

Abstract: Personas are often employed to guide large language model agents, yet their effectiveness in shaping strategic behavior in social dilemma settings remains uncertain. To address this, we examined the impact of persona prompts in an iterated Split or Steal game where persona-driven agents interacted with a Virtual Human (VH) controlled by a fixed prompt.

摘要： 人格设定（Personas）常被用于引导大语言模型智能体，但其在社会困境场景中塑造战略行为的有效性尚不明确。为了解决这一问题，我们研究了人格提示词在迭代式“平分或独吞”（Split or Steal）博弈中的影响。在该实验中，由人格驱动的智能体与一个由固定提示词控制的虚拟人（Virtual Human, VH）进行互动。

Agents were instantiated from four open models (Ministral 3:3b, phi4:14b, Gemma3:12b, and Gemma4:e4b) at two temperature settings (0.3 and 0.7) and deterministic decision with zero temperature, while the VH was powered by GPT 4.1 mini. Across 160 sessions of 15 rounds each conducted in European Portuguese, mutual Split outcomes dominated (roughly 74 percent of rounds), with exploitation occurring in fewer than 11 percent of rounds.

智能体基于四种开源模型（Ministral 3:3b、phi4:14b、Gemma3:12b 和 Gemma4:e4b）实例化，并设置了两种温度参数（0.3 和 0.7）以及零温度下的确定性决策，而虚拟人则由 GPT 4.1 mini 驱动。在以欧洲葡萄牙语进行的 160 场（每场 15 轮）实验中，双方“平分”的结果占据主导地位（约占总轮次的 74%），而出现“独吞”行为的比例不到 11%。

Model choice significantly influenced behavior: phi4 and Ministral 3:3b remained consistently cooperative across temperatures, whereas Gemma3:12b and Gemma4:e4b exhibited more varied strategies and outcomes. Analyses based on Big Five personality traits indicated that Prosocial and Principled personas were most consistently cooperative, while Analytical personas were more likely to exploit the VH.

模型选择对行为有显著影响：phi4 和 Ministral 3:3b 在不同温度设置下始终保持合作，而 Gemma3:12b 和 Gemma4:e4b 则表现出更多样化的策略和结果。基于“大五人格”特质的分析表明，“亲社会型”和“原则型”人格最为稳定地表现出合作倾向，而“分析型”人格则更有可能利用（剥削）虚拟人。

Topic analysis revealed that friendship-related dialogue aligns with Split decisions, whereas money and vengeance-related content is more prevalent in Steal outcomes; sentiment labels were predominantly neutral or happy and provided limited additional explanatory value. These findings characterize the interaction between persona prompts and model differences in repeated trust games and serve as a baseline for planned virtual reality studies involving human participants interacting with an embodied VH.

主题分析显示，与友谊相关的对话往往与“平分”决策一致，而涉及金钱和报复的内容在“独吞”结果中更为常见；情感标签主要为中性或快乐，提供的额外解释价值有限。这些发现刻画了人格提示词与模型差异在重复信任博弈中的相互作用，并为后续涉及人类参与者与具身虚拟人互动的虚拟现实研究提供了基准。