From Plausible Agents to Accountable Simulation

2026

Open publication workspace · Sign in to read the full PDF.

AI-generated summary

1) This technology report establishes a technical and governance framework for LLM-enabled agent-based modelling, addressing how to transition from plausible agent behaviors to accountable and validated social simulations.
2)
- Analyzes fourteen public LLM-agent projects and nine traditional agent-based platforms via source-code inspection to reveal hidden default behaviors, parsing errors, and reproducibility limitations.
- Establishes a five-tier evidence hierarchy to evaluate different roles of language models in simulations, ranging from face-validity plausibility to monitored decision support.
- Proposes comprehensive technical and governance recommendations—including full stack disclosure, robustness audits, and community participation—to ensure reliable use in public policy.

3) Tags: LLM, agent-based modelling, social simulation, AI governance, validation

Check the original publication for accuracy and context.