Generative Agent-Based Modeling (GABM)

We are working on a new approach to modeling complex social systems where generative AI is used to better represent human behavior.  We refer to this new paradigm of modeling as "generative agent-based modeling," GABM. In these models, individuals are "autonomous", empowered by generative AI. 

Our recent papers: 

1. Aoki, G., & Ghaffarzadegan, N. (2026). AI Agents as Policymakers in Simulated Epidemics. arXiv preprint arXiv:2601.04245. 

We embed a generative agent inside a system dynamics simulated world, tasked with policy making. The simulated world is the conventional SEIR and SEIRb model. In our systems-knowledge intervention, when the generative agent is provided with a short text prompt that explains epidemic feedback loops in a plain language, she end up saving many lives in the simulated world!

2. Nour, S.R., Ghaffarzadegan, N., Naugle, A., Godfrey, J. (2026) Simulating U.S. Presidents for a Friendly Chat: Applying Generative AI to Study Political History. In: Artificial Intelligence in Modeling and Simulation, Philippe J. Giabbanelli and Istvan David (eds). Series on Simulation Foundations, Methods and Applications. Springer, Cham: 2195-2817 

We explore how AI agents can be trained to act as specific historical individuals. As a demonstration, we simulate 60 U.S. presidents (1789–2025), trained based on their inaugural addresses, and then ask each simulated president what factors influence the economy. We validate the responses using independent LLMs asking the LLMs which president would have given this response, finding fairly satisfactory results. Having the response from all (simulated) presidents, we then visualize their mental models and compare their mental model distances (differences) in a network diagram. We also analyze how presidential thinking evolves over time. 

3. Nour, S. R., Ghaffarzadegan, N., Majumdar, A., & Hosseinichimeh, N. (2025). Birds of a feather: clustering mental models to explore how people think alike. Journal of the Operational Research Society, 1-17. 

4. Hosseinichimeh, N., Majumdar, A., Williams, R., & Ghaffarzadegan, N. (2024). From text to map: a system dynamics bot for constructing causal loop diagrams. System Dynamics Review, e1782.  

5. Ghaffarzadegan, N., Majumdar, A., Williams, R., & Hosseinichimeh, N. (2024) Large Language Models May Capture Human Preferences and Decisions: The Persona Hypothesis, Researchgate. DOI: 10.13140/RG.2.2.11782.18242

6. Ghaffarzadegan, N., Majumdar, A., Williams, R., & Hosseinichimeh, N. (2024). Generative agent‐based modeling: an introduction and tutorial. System Dynamics Review, 40(1), e1761.  

The concept of Generative Agent-Based Modeling as an approach that combines LLM with mechanistic models to represent decision making is established and demonstrated using a simple model of norm diffusion, where individuals act as quasi-autonomous agents.

7. Williams, R., Hosseinichimeh, N., Majumdar, A., & Ghaffarzadegan, N. (2023). Epidemic modeling with generative agents. arXiv preprint arXiv:2307.04986. 

Media coverage: 

• Daily Beast: ChatGPT Invented This Fake Epidemic. Can It Help Prevent the Next One?

• Daily Mail: Could ChatGPT save us from the next pandemic? Researchers are using the free AI to simulate future outbreaks

• Asharq Al-Awsat: Could AI Help Prevent Future Epidemics?

• Euronews: Disease X: How AI could help plan our response to future pandemics

8. Ongoing projects. 

• MarketX: This project is about modeling customer choices with different demographics and preferences. The project is finished, and the client is already using the outcomes in a large scale. The program is confidential and the IP is purchased by the client. 

• AgentX

• CommunityX