Role Steering of Language Models for Social Simulations

2026-07-09

AI Summary

This study addresses the challenge of maintaining behavioral consistency in language model agents during social simulations and the absence of effective pre-deployment screening mechanisms. The authors propose a role-screening pipeline based on activation steering, which enables fine-grained control by defining role profiles, extracting role-specific activation directions, scanning steering coefficients, and evaluating behavioral consistency. For the first time, experiments across 275 roles demonstrate that steering intensity should be individually optimized per role rather than uniformly applied. Using the OLMo-3-7B-Instruct model with GPT-4.1-mini for auxiliary evaluation, the proposed method significantly outperforms baselines in role consistency (63.2 vs. 41.1) while preserving high lexical diversity. Notably, performance declines with increased steering strength for 38 roles, highlighting the necessity of role-adaptive calibration.

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. 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. 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 https://anonymous.4open.science/r/anonymous-research-code-5F03/.

Problem

• role steering
• language models
• social simulations
• role-conditioned behavior
• agent alignment

Innovation

• activation steering
• role-conditioned agents
• social simulations
• lexical diversity
• role-profile alignment