The Universitat Pompeu Fabra team analyzes how power dynamics and working conditions affect disease burden to translate model findings into actionable tools for health equity. The global approach to health surveillance is experiencing a major shift with the launch of the AIBEM project, an ambitious initiative designed to redefine how the scientific community predicts the spread of infectious diseases. Led by a consortium including the University of Zaragoza, Universidad Carlos III de Madrid, and Universitat Pompeu Fabra, this project recently secured significant funding through the 2026 Health Research Call by the “la Caixa” Foundation. By integrating advanced technology with public health strategy, the initiative aims to create a more resilient framework for managing biological threats. Traditional epidemiological models have historically focused on the biological movement of pathogens through contact networks, often overlooking the unpredictable nature of the people involved. The AIBEM project addresses this gap by placing the human factor at the center of its research. Instead of viewing behavior as a static variable, this project models how social determinants—such as public trust in authorities, the spread of misinformation, and socioeconomic status—actively influence the trajectory of an epidemic. This transition represents a fundamental departure from the purely mathematical frameworks of the past, signaling a new era where sociology and computer science converge to protect global populations more effectively.
Human-Centric Forecasting: Bridging the Gap in Mathematical Logic
The COVID-19 pandemic revealed a significant weakness in existing scientific frameworks: while researchers could track the virus, they struggled to predict how the public would respond to mandates. Compliance with mask-wearing or vaccination was never uniform, as it was constantly influenced by shifting risk perceptions and institutional trust. AIBEM seeks to move beyond these limitations by treating human behavior as a dynamic, evolving element that is just as critical to the simulation as the virus itself. For decades, the scientific community relied on models that assumed humans behaved like particles in a gas—predictable and uniform. However, real-world data suggested that people react to information, social pressure, and economic constraints in ways that drastically alter the infection rate of a pathogen. By acknowledging that a lockdown is only as effective as the community’s willingness or ability to comply, the AIBEM project builds a bridge between biological reality and social psychology. This nuanced approach ensures that when a new pathogen emerges, the models used to fight it will reflect a society of thinking, feeling individuals rather than a collection of static data points.
To achieve this level of realism, the project utilizes Generative Agent-Based Models (GABMs), which leverage the same technology found in Large Language Models. These models create autonomous computational agents that possess their own identities, goals, and reasoning capabilities. These digital entities do not just represent data; they interact socially, influence one another, and develop social norms, allowing researchers to observe how psychological and social pressures might impact the success of a public health campaign. Unlike traditional agent-based models that followed simple “if-then” rules, GABMs can simulate complex decision-making processes, such as an individual weighing the risk of infection against the need to earn a daily wage. These agents can digest news reports, communicate with other agents, and change their behavior based on the prevailing sentiment in their virtual community. This creates a playground for health officials to test communication strategies, identifying which messages might inadvertently spark skepticism and which will foster collective action. The integration of generative AI allows for the simulation of billions of interactions that would be impossible to map manually, providing a high-fidelity look at the social fabric of an epidemic.
Interdisciplinary Synergy: Collaborative Expertise Across Academic Institutions
The success of the initiative relies on a collaborative framework where each institution contributes a specific expertise to the overarching goal of pandemic resilience. The University of Zaragoza handles the primary development of these generative agents, utilizing its sophisticated BIFI Institute supercomputing infrastructure to run massive simulations that require significant processing power. By hosting the project’s computational core, Zaragoza ensures that the complex reasoning of thousands of AI agents can be processed simultaneously without lag, providing real-time data to researchers. Meanwhile, Universidad Carlos III de Madrid focuses on behavioral validation, which serves as the crucial check against artificial hallucination or unrealistic AI logic. Their team conducts large-scale experiments with real human participants via the IBSEN platform to ensure the AI agents act in ways that accurately reflect genuine human decision-making processes. This iterative process of building a model and then testing it against real human behavior creates a feedback loop that constantly improves the accuracy of the simulations. This combination of raw computing power and behavioral science ensures that the resulting tools are both technically robust and sociologically sound.
Building on the technical foundation provided by Zaragoza and Madrid, the team at Universitat Pompeu Fabra focuses on the social and economic dimensions of health. Their researchers analyze how power dynamics and working conditions contribute to the overall disease burden, moving the model beyond pure mathematics and into the realm of social advocacy. By translating complex data into actionable policy tools, they ensure that the findings can be used to protect the most vulnerable members of society who often bear the brunt of public health crises. This focus on ecosocial determinants acknowledges that a person’s ability to isolate or seek medical care is often dictated by their employment status or housing situation. The UPF contribution is vital for turning the AIBEM project from a mere academic exercise into a practical instrument for health equity. They work to identify specific points of failure in the social safety net where a virus is most likely to take hold and spread rapidly due to neglect or lack of resources. By integrating these factors into the generative agents, the models can simulate a society where inequality is a known variable, allowing for the creation of interventions that address the root causes of transmission.
Social Responsibility: Integrating Equity and Ethical AI Standards
A defining feature of the AIBEM project is its focus on socioeconomic equity, acknowledging that epidemics are never purely biological events but are instead deeply intertwined with the material conditions of the population. The models incorporate detailed census data and labor surveys to account for variables like income, occupation, and neighborhood characteristics, which were often ignored in older transmission matrices. This allows for simulations that show how specific groups—such as essential workers who cannot work from home or families living in overcrowded urban areas—face significantly higher risks of infection and severe outcomes. By mapping these disparities in a virtual environment, health authorities can design targeted interventions that reduce inequalities rather than making them worse through blanket policies. For instance, instead of a city-wide lockdown that disproportionately harms low-income workers, the model might suggest subsidized transportation or localized health clinics as a more effective and just solution. This data-driven approach to social justice ensures that public health strategy is not just about stopping a virus, but about protecting all members of society equally.
Governance and ethics are also central to the project’s mission, ensuring that the AI does not inadvertently reinforce social prejudices or generate biased outcomes. The project follows the European AI Act, incorporating rigorous bias audits and human oversight into its development process to maintain the highest standards of digital responsibility. By proactively identifying and mitigating potential biases in the Large Language Models used for agent reasoning, the consortium ensures that the simulations do not stigmatize specific communities or overlook certain demographic nuances. Furthermore, by adhering to open science principles, the consortium plans to make its models, datasets, and validation protocols accessible to the global scientific community to foster transparency and international cooperation. This commitment to FAIR principles means that the breakthroughs made by the AIBEM project can be utilized by researchers in other countries, potentially saving lives far beyond the borders of Spain. In an era where trust in technology is paramount, this transparent approach serves to build public confidence in AI-driven health policies while allowing external experts to scrutinize and build upon their collective work.
Future-Proofing Public Health: Transforming Preparedness Through Virtual Simulations
The ultimate goal of the AIBEM project is to provide a comprehensive toolkit for health authorities across Europe, transforming how governments prepare for and react to emerging health threats. This includes a suite of validated behavioral agents and coupled models that can quantify how factors like vaccine hesitancy or pandemic fatigue change transmission rates over time. By providing sophisticated decision-support software, the project allows policymakers to test the impact of various interventions, such as a localized lockdown or a targeted financial subsidy, in a virtual environment before implementing them in the real world. This pre-testing capability is crucial for avoiding the costly and sometimes traumatic trial-and-error approach that characterized previous global health emergencies. These tools are designed to be user-friendly, allowing public health officials without deep coding expertise to run scenarios and visualize the potential outcomes of their choices. The ability to see how a specific policy will ripple through different socioeconomic layers of a city provides a level of foresight that was previously impossible. This technological bridge between research and governance ensures that the most recent scientific insights are immediately available to those with the power to enact change.
Looking back at the development of the AIBEM project, it established a new benchmark for how technology served humanity during times of crisis. The consortium successfully demonstrated that the integration of generative AI into epidemiological models provided the necessary clarity to manage complex social behaviors that once baffled health authorities. This initiative moved the scientific community closer to a future where biological threats were met with a coordinated, equitable, and highly informed response. Health departments that adopted these generative tools found themselves better equipped to combat misinformation and maintain public trust, as their policies were grounded in realistic simulations of human nature. The actionable next steps for the global health community included the standardization of these agent-based models across international borders to ensure a unified defense against transboundary pathogens. Researchers emphasized that the continued investment in high-performance computing and behavioral data collection remained essential for the longevity of these systems. Ultimately, the project’s legacy was defined by its ability to prioritize the human element within a digital framework, ensuring that the pursuit of health was always synonymous with the pursuit of social justice and shared human resilience.
