An Infectious Disease Spread Simulation Based on Large Language Model Decision Making

2019-07-26 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Public Health & Epidemiology · Depth: Expert, extended

Summary

Researchers developed a spatially grounded, agent-based simulation framework that integrates Large Language Models (LLMs) to model individual decision-making during infectious disease outbreaks. This framework replaces traditional rule-based mechanisms with LLM-generated decisions regarding self-reported influenza-like illness, using a synthetic population based on real-world census data for San Francisco and Atlanta. The simulation explores three scenarios: independent reasoning, household influence, and message framing. Results indicate that income and education are the primary drivers of symptom reporting rate variation, with geography, LLM model choice, and message framing having smaller but consistent impacts. For instance, under the independent scenario, mean reporting rates were 65.4% in Atlanta and 64.7% in San Francisco. The framework's source code is publicly available, and it utilizes open-source LLMs like Meta Llama-3-8B-Instruct and Google Gemma-2-9B-IT.

Key takeaway

For AI Scientists and Research Scientists designing public health simulations, you should consider integrating Large Language Models to capture nuanced, demographically sensitive behavioral dynamics. Your models will better reflect real-world disparities, such as those driven by income and education, but be mindful of LLM choice and prompt sensitivity. Validate your LLM-driven agent behaviors against real-world data to ensure predictive accuracy and avoid reinforcing biases.

Key insights

LLMs can simulate nuanced human health behaviors in spatially-grounded agent-based models, revealing demographic disparities.

Principles

• LLM-driven agents capture behavioral heterogeneity.
• Demographic factors strongly predict health reporting.
• Contextual framing influences LLM-generated decisions.

Method

The framework extends a Java-based simulator, incorporating an SEIR disease model and LLM-based decision-making. It pre-generates decisions for demographic combinations using LLMs, storing them in a decision bank for runtime retrieval.

In practice

• Use LLMs to model health-seeking behaviors.
• Incorporate census data for realistic agent demographics.
• Test message framing for public health interventions.

Topics

• Large Language Models
• Agent-Based Modeling
• Infectious Disease Simulation
• Public Health Interventions
• Demographic Bias
• Health Behavior Modeling

Code references

• cruiseresearchgroup/disease-simulator-LLM_agent

Best for: AI Scientist, Research Scientist, AI Engineer

Related on AIssential

• An Infectious Disease Spread Simulation Based on Large Language Model Decision Making — LLMs can simulate human decision-making in spatially-grounded infectious disease models, revealing key socioeconomic drivers.
• When eight popular large language models (LLMs) answer common parent questions about early maxillary expansion, how reliable are they—and how readable is what they say? — LLM performance in healthcare varies significantly across accuracy, comprehensiveness, and readability, with no single model excelling in all.
• Agentifying Patient Dynamics within LLMs through Interacting with Clinical World Model — Integrating a Clinical World Model with LLMs via agentic training improves sepsis treatment recommendations and safety.
• An Infectious Disease Spread Simulation Based on Large Language Model Decision Making — LLMs can simulate human behavior in spatially-grounded disease spread models, revealing demographic drivers.
• Position: AI Agents Are Not (Yet) a Panacea for Social Simulation — LLM agents for social simulation require explicit environment modeling and rigorous evaluation beyond mere plausibility.
• Modeling Pathology-Like Behavioral Patterns in Language Models Through Behavioral Fine-Tuning — Consistent behavioral fine-tuning can induce stable, pathology-like biases in LLM generative distributions and action selection.

Open in AIssential →

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