Programmable Cognitive Bias in Social Agents

2025-09-16

AI Summary

Traditional LLM-based social simulation relies on implicit natural-language descriptions of agent behavior, resulting in inconsistent cross-model behavior and poor fidelity in capturing fine-grained cognitive differences. To address this, we propose CoBRA—a novel toolkit that explicitly encodes cognitive biases into social agent control. CoBRA introduces quantifiable cognitive bias indices derived from canonical social psychology experiments and implements a model-agnostic behavioral regulation engine. This engine leverages prompt engineering and feedback-driven behavioral regularization to achieve precise, reproducible alignment of agent behavior with targeted biases. Experiments demonstrate that CoBRA consistently induces and modulates specific cognitive biases across diverse large language models, significantly improving behavioral consistency, simulation credibility, and cross-model reproducibility. By enabling programmable, bias-aware social simulation, CoBRA establishes a new paradigm for rigorous human-AI interaction research and controllable agent-based social modeling.

Technology Category

• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior
• Multiagent Systems: Agent-Based Simulation and Emergent Behavior
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

This paper introduces CoBRA, a novel toolkit for systematically specifying agent behavior in LLM-based social simulation. We found that conventional approaches that specify agent behaviors through implicit natural language descriptions cannot yield consistent behaviors across models, and the produced agent behaviors do not capture the nuances of the descriptions. In contrast, CoBRA presents a new approach to program agents' cognitive biases explicitly, by grounding agents' expected behaviors using classic social science experiments. CoBRA has two components: (1) Cognitive Bias Index that measures the cognitive bias of a social agent, by quantifying the agent's reactions in a set of validated classical social science experiments; (2) Behavioral Regulation Engine that aligns the agent's behavior to demonstrate controlled cognitive bias. We evaluated CoBRA as an HCI toolkit through demonstration and technical benchmarks. Our results suggest that CoBRA can precisely program the cognitive bias demonstrated in a social agent in a model-agnostic manner.

Problem

• Specifying consistent agent behaviors in LLM simulations
• Measuring cognitive bias through social science experiments
• Programming controlled cognitive bias in model-agnostic manner

Innovation

• Programs cognitive bias explicitly using social experiments
• Measures bias via Cognitive Bias Index quantification
• Aligns behavior with Behavioral Regulation Engine control