Intel / article

Role Steering of Language Models for Social Simulations

Source: arxiv.org
Author: Isaac Song, Mohammed Rehan Parwani, Glenn Matlin, Emile Anand, Akhil Theerthala, Arjun Chatterjee, Maria Kostylew, Yonadav G. Shavit, Sebastien Krier, Mark Riedl
Date: Published Aug 4, 2026

Context

Making a simulated AI character more strongly follow a role can make some roles score worse. Researchers tested adjustments to internal model activations across 275 roles using OLMo-3-7B-Instruct, trying four strengths for each role. Most roles improved as steering increased, but 38 declined on every measured dimension. The proposed workflow therefore screens each role and setting before including it in a simulated population. Its scores measure agreement with model-generated role references, judged by GPT-4.1-mini; they establish performance against those references rather than fidelity to actual human behavior. Research abstract.

Key takeaways · AI-distilled

Role-specific activation directions beat the prior persona-vector assistant-axis control decisively — 63.2 vs 41.1 mean judged role- across the tested grid on OLMo-3-7B-Instruct with 275 roles.
Role directions also preserve lexical diversity, while the assistant-axis control drops sharply at larger steering coefficients.
38 of 275 roles (~14%) degrade across all six measured dimensions at every steering strength — so coefficients must be chosen per role via screening, never deployed as one uniform high-strength setting.
The reusable recipe: define a role profile, extract its direction, sweep four coefficients, judge alignment against the profile, then pass or flag each configuration before it enters a simulation.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
alignment — The work of making AI systems actually pursue what their builders and users intend, rather than something subtly or dangerously different.

Key quotes

“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.”

“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.”

Read the source arxiv.org