Researchers selected in the 2026 Health Research Call by the “la Caixa” Foundation have launched the AIBEM project to model human behavior using generative artificial intelligence. Backed by €970,139 in funding, the consortium aims to track epidemic spread by simulating public trust, socioeconomic inequalities, and compliance with health guidelines.
In Plain English: The Clinical Takeaway
- Generative Agent-Based Models: Computer simulations where individual digital agents are powered by large language models, allowing them to exhibit human reasoning and behavioral adaptation during health crises.
- Socioeconomic Stratification: Incorporating real-world experimental data into epidemiological projections to ensure that disease burdens and policy impacts reflect material conditions rather than homogenous population assumptions.
- Prospective Policy Evaluation: Testing the potential societal and epidemiological outcomes of public health interventions—such as mobility restrictions or vaccination campaigns—before authorities implement them in the real world.
The Computational Shift in Modeling Human Behavior
For two decades, science has allowed researchers to accurately model the dynamics of pathogen spread through contact networks, but human behavior during health crises has remained difficult to predict. The AIBEM project addresses this limitation by using artificial intelligence to replicate real-world behavioral dynamics.
The consortium unites multiple institutions, including the University de Zaragoza, Universidad Carlos III de Madrid (UC3M), and Universitat Pompeu Fabra (UPF). According to Yamir Moreno—who serves as the project’s principal investigator, Director of the Institute for Biocomputation and Physics of Complex Systems (BIFI), and a member of the Department of Theoretical Physics at the University de Zaragoza—the emergence of generative AI reshapes the paradigm by providing autonomous computational agents with mutual influence and reasoning faculties.
Validating Generative Agents Against Human Experiments
Building reliable computational agents requires rigorous empirical testing against actual human responses. To achieve this, researchers at UC3M utilize the IBSEN platform to conduct behavioral experiments with socioeconomically diverse participants. Anxo Sánchez, a researcher on the project from UC3M’s Interdisciplinary Group of Complex Systems (GISC), explained that validating generative agents requires comparing their outputs against standard experiments and modeling epidemic-related scenarios.
This methodology builds upon parallel academic frameworks, such as the Generative Adaptive Behavioral Layer for Epidemics (GABLE) detailed in recent preprint literature (arXiv:2608.29535). GABLE demonstrates how large language models can infer behavioral adjustments to epidemic conditions and translate them into age-structured contact matrices. In comparative testing, these LLM-generated matrices outperformed mobility-driven matrices derived from real-world mobility data, particularly over longer forecasting horizons.
| Model Component | Traditional Epidemiological Models | AIBEM Generative Agent Framework |
|---|---|---|
| Human Behavior | Treated using predefined response functions or economic frameworks | Dynamic agents powered by large language models |
| Socioeconomic Factors | Generally aggregated | Integrated via experimental data with explicit socioeconomic diversity |
| Policy Evaluation | Retrospective analysis of implemented measures | Prospective simulation before policy implementation |
Prioritizing Equity in Public Health Interventions
Epidemics are deeply conditioned by living and working conditions and power dynamics. Recognizing that advanced prediction tools are insufficient if they fail to incorporate structural drivers, the UPF team focuses on ecosocial determinants and health inequalities. Joan Benach—who is the Director of the Health Inequalities Research Group – Employment Conditions Network (GREDS-EMCONET) and Co-Director of the JHU-UPF Public Policy Center at Universitat Pompeu Fabra (UPF)—stressed that the principal objective is formulating equitable public health strategies that mitigate rather than amplify existing social gaps.
To ensure structural vulnerabilities are incorporated, the project utilizes behavioral agents validated against real human experimental data. Furthermore, all models, datasets, and validation protocols will be published open-access following FAIR principles.
Contraindications & When to Consult a Doctor
Computational epidemiological models and artificial intelligence tools are designed for macroscopic public health planning, institutional decision-making, and policy evaluation. They do not constitute diagnostic medical devices, personal health advice, or individual clinical protocols. Individuals experiencing symptoms of an infectious disease should never rely on public health modeling software or artificial intelligence platforms for personal medical triage. Always consult a qualified healthcare professional, local public health authority, or clinical provider for personalized medical diagnosis, testing, vaccination schedules, and treatment guidance during an outbreak.
Future Trajectory and Institutional Integration
Selected as one of 32 projects from 800 proposals, AIBEM secures €970,139 over a three-year timeline within a broader €25.7 million funding allocation by the “la Caixa” Foundation. Among the projected outputs are paired epidemiological simulations, behavioral agents tested against empirical human data, and an analytical toolkit tailored for European and Spanish health authorities comprising simulation software, ethical standards, and policy memos.
References
- University de Zaragoza & Universidad Carlos III de Madrid. AI-Driven Behavioural Epidemic Models for Intervention Design (AIBEM). La Caixa Foundation Health Research Call, 2026.
- Integrating adaptive human behavior into epidemic models with large language models. arXiv preprint arXiv:2608.29535, 2026.
Disclaimer: This article is published exclusively on Archyde.com for informational and public health reporting purposes. It does not replace professional medical advice, diagnosis, or treatment.
