Back to AI Research

AI Research

ScioMind: Cognitively Grounded Multi-Agent Social S... | AI Research

Key Takeaways

  • ScioMind: Cognitively Grounded Multi-Agent Social Simulation with Anchoring-Based Belief Dynamics and Dynamic Profiles ScioMind is a new simulation framework...
  • Large language model (LLM)-based multi-agent simulation offers a powerful testbed for studying social opinion dynamics.
  • Yet current approaches often adopt two contrasting methods: either relying on fixed update rules with limited cognitive grounding or delegating belief change largely to unconstrained LLM interaction.
  • We introduce ScioMind, a cognitively grounded simulation framework that bridges these paradigms by combining structured opinion dynamics with LLM-based agent reasoning.
  • We evaluate ScioMind on multiple case studies in a real-world policy debate scenario.
Paper AbstractExpand

Large language model (LLM)-based multi-agent simulation offers a powerful testbed for studying social opinion dynamics. Yet current approaches often adopt two contrasting methods: either relying on fixed update rules with limited cognitive grounding or delegating belief change largely to unconstrained LLM interaction. We introduce ScioMind, a cognitively grounded simulation framework that bridges these paradigms by combining structured opinion dynamics with LLM-based agent reasoning. ScioMind integrates three key components: 1) a memory-anchored belief update rule that modulates susceptibility to influence via personality-conditioned anchoring strength; 2) a hierarchical memory architecture that supports persistent, experience-driven belief formation; and 3) dynamic agent profiles derived from a corpus-grounded retrieval pipeline, enabling heterogeneous personalities, rationales, and evolving internal states. We evaluate ScioMind on multiple case studies in a real-world policy debate scenario. Across metrics including polarisation, diversity, extremization, and trajectory stability, the proposed components consistently yield improvements in behavioural realism. In particular, dynamic profiles increase opinion diversity, memory and reflection reduce unstable oscillation, and anchoring induces persistent belief trajectories that better align with patterns reported in political psychology. These results suggest that our cognitively grounded design provides a novel solution to LLM-based social simulation that improves both stable and behavioural realism

ScioMind: Cognitively Grounded Multi-Agent Social Simulation with Anchoring-Based Belief Dynamics and Dynamic Profiles
ScioMind is a new simulation framework designed to study how opinions form and change within social networks. While previous methods for simulating social behavior often rely on either rigid mathematical rules or unconstrained AI interactions, ScioMind bridges these two approaches. It combines structured social science models with the reasoning capabilities of Large Language Models (LLMs) to create a more realistic testbed for observing how beliefs persist, evolve, and polarize during policy debates.

How the Framework Works

The core of ScioMind is a "memory-anchored" belief update system. Inspired by cognitive psychology, the framework recognizes that people often hold onto their initial beliefs—or "anchors"—even when presented with new information. In this model, an agent’s resistance to changing their mind is determined by their specific personality traits, such as openness or conscientiousness. By using a four-layer memory architecture (episodic, semantic, procedural, and reflection), the system allows agents to maintain a consistent sense of their own history, which in turn influences how they process new social interactions.

Creating Realistic Agents

Rather than using static, pre-defined roles, ScioMind generates dynamic agent profiles. The system pulls from a corpus of real-world data to build diverse identities, complete with unique interests, backgrounds, and psychological profiles based on the "Big Five" personality traits. These agents are then placed into social networks where their relationships—such as friendships or professional connections—are determined by shared interests, group affiliations, and their specific stances on issues. This allows the simulation to capture the nuance of human social circles rather than treating agents as isolated data points.

Improving Simulation Realism

The researchers tested ScioMind against real-world policy debate scenarios to see how well it mimics human behavior. The results showed that the framework significantly improved the realism of the simulations. Specifically, the use of dynamic profiles helped maintain a healthy diversity of opinions, while the memory and reflection components prevented the "unstable oscillation" often seen in other models. By incorporating anchoring, the agents exhibited the kind of persistent belief trajectories that are commonly observed in political psychology, avoiding the tendency of other systems to force agents toward a neutral consensus too quickly.

Key Considerations

ScioMind represents a shift toward "cognitively grounded" social simulation. By explicitly modeling how personal experience and personality shape the way individuals resist or accept influence, the framework provides a more transparent and interpretable way to study social dynamics. This approach allows researchers to identify exactly what drives an agent’s opinion change, offering a more robust tool for understanding the complex, often stubborn nature of human belief formation in the digital age.

Comments (0)

No comments yet

Be the first to share your thoughts!