LLM-powered reasoning in agent-based modeling

2025-08-21 · Source: cs.MA updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Robotics & Autonomous Systems · Depth: Advanced, extended

Summary

The Hybrid Agent-based and Language-driven Epidemic (HALE) modeling framework introduces a novel approach to enhance agent-based modeling (ABM) by integrating Large Language Models (LLMs) to predict dynamic human decision-making in real-time scenarios. Traditionally, ABMs rely on static data, failing to adapt to evolving conditions like epidemics. HALE addresses this by coupling ABM for disease dynamics with LLMs for weekly updates to agent mobility patterns, enabling near real-time forecasting. As a proof-of-concept, HALE simulated COVID-19 in Salt Lake County, UT, from September 2020 to February 2022, involving 1.1 million agents. The framework, utilizing Llama 3.1 with a temperature of 0.2, better captured the observed epidemic peak and size, with total infections 2.21 times higher than symptomatic observations, aligning with estimates of asymptomatic cases. In contrast, ABM-only simulations tended to overestimate the epidemic's effects.

Key takeaway

For public health analysts and epidemiological modelers developing population-scale digital twins, the HALE framework offers a critical advancement. By integrating LLMs to dynamically model human behavior, your simulations can achieve greater accuracy in forecasting epidemic peaks and sizes, especially when traditional ABMs rely on static data. This improved fidelity in predicting real-world responses to outbreaks can inform more effective policy decisions and resource allocation, moving beyond retrospective analysis to near real-time adaptability.

Key insights

LLMs dynamically update human behavior in agent-based epidemic models, improving real-time forecasting accuracy.

Principles

• Hybrid ABM-LLM models enhance real-time epidemic forecasting.
• LLMs infer group-level behavioral changes based on context.
• Spatial and demographic grouping enables scalable LLM integration.

Method

The HALE framework couples ABM for disease dynamics with LLMs for weekly updates to agent mobility patterns, using structured outputs and batch processing for efficiency.

In practice

• Group agents by geography and demographics for LLM inference.
• Use a balanced LLM temperature (e.g., 0.2) for behavioral prediction.
• Integrate LLMs as GPU-backed inference servers with ABM clients.

Topics

• Large Language Models
• Agent-Based Modeling
• Epidemic Modeling
• COVID-19
• Human Behavior Prediction
• Spatiotemporal Networks

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Prospective multi-pathogen disease forecasting using autonomous LLM-guided tree search — LLM-guided tree search autonomously generates and optimizes disease forecasting models, matching human expert performance.
• Bridging the Last Mile of Time Series Forecasting with LLM Agents — LLM agents can bridge the gap between statistical time series forecasts and business-ready predictions by integrating weakly structured contextual information.
• An Infectious Disease Spread Simulation Based on Large Language Model Decision Making — LLMs can simulate nuanced human health behaviors in spatially-grounded agent-based models, revealing demographic disparities.
• Design a Reliable LLM-Integrated Interface for Mortality Forecasting — LLMs can democratize complex mortality forecasting by translating natural language into structured, statistically valid configurations.
• Harnessing Generalist Agents for Contextualized Time Series — TimeClaw's agentic harness framework enables LLMs to perform robust, contextualized time series reasoning by providing native temporal runtime support.
• Team MKC at CLPsych 2026: Capturing and Characterizing Mental Health Changes through Social Media Timeline Dynamics — An LLM-based pipeline unifies post-level assessment and user-level temporal modeling for mental health analysis from social media.

Open in AIssential →

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