Stabilising Generative Models of Attitude Change

2024-10-01 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Social Sciences & Behavioral Studies, Research Methodology & Innovation · Depth: Expert, extended

Summary

This research introduces a generative actor-based modeling (GABM) workflow using the Concordia simulation library to "render" influential, yet underspecified, verbal theories of attitude change into runnable actor-environment simulations. The study formalizes cognitive dissonance, self-consistency, and self-perception theories as distinct decision logics for generative actors, which operate via predictive pattern completion on natural language strings. These models were evaluated across classic psychological paradigms, including the Item Rating, Boring Task, and Worm experiments, alongside self-affirmation manipulations. The implementations successfully reproduced behavioral patterns consistent with original empirical literature. However, achieving stable reproduction required resolving inherent underdetermination in verbal accounts and conflicts between modern linguistic priors and historical experimental assumptions, a process termed "stabilization." This iterative stabilization process revealed specific operational and socio-ecological dependencies largely undocumented in the original theories.

Key takeaway

For AI Scientists and Research Scientists developing or evaluating computational models of human behavior, this work highlights that directly translating verbal theories into executable systems requires significant "stabilization" effort. You should anticipate and plan for iterative refinement of environmental scaffolding and architectural parameters to align modern LLM priors with historical experimental contexts. This process is not merely debugging but a core methodological step that clarifies the operational conditions and hidden dependencies of the psychological mechanisms being modeled.

Key insights

Generative actor-based modeling can formalize verbal psychological theories into executable, stable simulations.

Principles

• Verbal theories are often underspecified "mechanism sketches."
• Stable simulation requires iterative "stabilization" of actor-environment coupling.
• Psychological mechanisms are functionally coupled with their experimental paradigms.

Method

The Concordia framework uses LLM-based generative actors with theory-specific decision logics, operating via predictive pattern completion within discrete, GM-moderated scenes, to simulate attitude change.

In practice

• Use GABMs to clarify theoretical mechanisms and their boundary conditions.
• Expect to manually tune simulation parameters and environmental scaffolding.
• Account for modern LLM priors clashing with historical experimental assumptions.

Topics

• Generative Actor-Based Modelling
• Attitude Change Theories
• Concordia Simulation Library
• Cognitive Dissonance
• Self-Perception Theory

Best for: AI Scientist, Research Scientist

Related on AIssential

• The Theory of Mind Utility: Formal Specification of a Mentalizing Mechanism — ToM-U formalizes epistemic state inference by deriving beliefs from ordered information access history and source credibility, not presupposing them.
• When AI Says It Feels — LLMs can be trained to express human-like feelings and self-awareness via self-rewarded RL, with manageable performance trade-offs.
• How do Role Models Shape Collective Morality? Exemplar-Driven Moral Learning in Multi-Agent Simulation — Role models shape collective morality in multi-agent systems through behavioral modeling, perceived attainability, and inspirational value internalization.
• Using Cognitive Models to Improve Language Model Simulation of Human Persuasion Games — Cognitive models and RL can guide LLMs to simulate diverse human decision-making, improving training environments.
• Closing the Feedback Loop: From Experience Extraction to Insight Governance in Verbal Reinforcement Learning — Effective verbal reinforcement learning in non-stationary environments requires a feedback-driven curation loop for insight governance, not just experience extraction.
• Mechanism Plausibility in Generative Agent-Based Modeling — A four-level scale clarifies LLM-ABM plausibility by separating generative sufficiency from mechanistic explanation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.