Role Steering of Language Models for Social Simulations

语言模型在社会模拟中的角色引导 (Role Steering)

Abstract: Social simulations built from language-model agents need role-conditioned behavior that can be checked before agents are placed into a simulated population. We introduce an activation-steering screening workflow for role-conditioned agents: define a role profile, extract a role-specific direction, sweep four steering coefficients, evaluate role-profile alignment, and pass or flag each candidate configuration.

摘要： 基于语言模型智能体构建的社会模拟，需要一种在智能体进入模拟群体前即可进行验证的角色条件行为。我们引入了一种针对角色条件智能体的激活引导筛选工作流：定义角色画像、提取角色特定方向、扫描四个引导系数、评估角色画像对齐度，并对每个候选配置进行通过或标记。

On OLMo-3-7B-Instruct, we apply the workflow to a mixed 275-role inventory with 228 role-agnostic questions, GPT-4.1-mini prompted role references, and GPT-4.1-mini judges. Role-specific directions receive higher judged role-profile alignment than an assistant-axis directional control from prior persona-vector work, with mean overall scores of 63.2 versus 41.1 across the tested grid.

在 OLMo-3-7B-Instruct 模型上，我们将该工作流应用于包含 275 个角色的混合清单，并结合 228 个与角色无关的问题、GPT-4.1-mini 提示的角色参考以及 GPT-4.1-mini 评估员进行测试。结果显示，与先前人格向量研究中的“助手轴”方向控制相比，角色特定方向获得了更高的角色画像对齐度评分，在测试网格中的平均总分分别为 63.2 分和 41.1 分。

They also preserve high lexical diversity, while the control drops sharply at larger coefficients. The role-level screen is the main practical output: most roles improve as steering increases, but 38 roles decline across all six measured dimensions, showing why simulation builders should choose coefficients per role rather than deploy a uniform high-strength setting. We make our code and evaluation artifacts available at this https URL.

此外，该方法保持了较高的词汇多样性，而对照组在系数较大时词汇多样性急剧下降。角色层面的筛选是本研究的主要实际产出：大多数角色随着引导强度的增加而表现提升，但有 38 个角色在所有六个测量维度上均出现下降，这表明模拟构建者应针对不同角色选择相应的系数，而非采用统一的高强度设置。我们已将代码和评估工件发布在指定链接中。