The project, titled AI-Driven Behavioural Epidemic Models for Intervention Design (AIBEM), has been selected in the 2026 Health Research Call by the “la Caixa” Foundation, one of the most competitive biomedical research funding calls in Spain and Portugal. AIBEM was one of just 32 projects chosen from 800 submitted proposals, a success rate of roughly 4 percent, and will receive €970,139 over three years. In total, the foundation will award €25.7 million to the projects selected in this edition. The consortium’s ambition is to build generative agent-based models, or GABMs, in which every simulated individual is powered by a large language model—the same underlying technology behind platforms such as ChatGPT or Gemini—configured with a specific personal profile, memory, goals, and tools.

“For two decades, science has allowed us to accurately model the dynamics of pathogen spread through contact networks. However, the true challenge has always been the unpredictability of our own nature. Capturing the complexity of human behavior during a health crisis remains an unresolved challenge. The rise of generative AI changes the rules of the game in our favor: by equipping autonomous computational agents with reasoning capabilities and mutual influence, we can transform epidemiological models from mere mathematical projections into a faithful reflection of the human factor on a societal scale,” said Yamir Moreno, 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 of Zaragoza.

The technical core of AIBEM represents what the researchers describe as a paradigm shift. In conventional epidemiological modeling, human behavior is treated as exogenous—something imposed on the model from outside rather than emerging from it. Compliance rates are assumed, contact patterns are fixed, and the subtle feedback loops between policy, perception, and action are lost. In a generative agent-based model, by contrast, each simulated person can interact with others, make decisions, set expectations, and generate social norms. Trust in health authorities can erode or strengthen; misinformation can propagate through a simulated community; socioeconomic inequalities can shape who isolates, who keeps working, and who gets sick. The researchers say these agent responses will be systematically validated against experiments involving real human participants, a step they consider essential to keeping the simulations anchored in observed behavior rather than plausible-sounding fiction.

“UC3M’s main contribution is its extensive experience in designing and executing behavioral experiments with human participants,” said Anxo Sánchez, a researcher on the project from UC3M’s Interdisciplinary Group of Complex Systems (GISC). “This capability is key, as validating generative agents requires not only comparing their behavior against standard experiments, but also modeling epidemic-related scenarios and designing experiments that faithfully replicate them—many of which have not been addressed before and will be crucial to the project’s success.”

The work is divided into three main packages distributed across the consortium. UC3M will run behavioral experiments with socioeconomically diverse participants using the IBSEN platform, developed within the framework of the European H2020 project of the same name. The University of Zaragoza will build the generative agents, replicate human behavior, and integrate them into epidemiological models, drawing on BIFI’s supercomputing infrastructure—the CESAR system—and the accumulated expertise of the Complex Networks and Systems Group (COSNET) in mathematical and computational epidemic modeling. UPF will translate the results into decision-support tools for health authorities, including an explicit assessment of how the costs and benefits of each intervention are distributed across society. That distributional analysis is not an add-on, the team insists, but a defining feature of the project.

“Epidemics are not purely biological phenomena: their spread and consequences are deeply conditioned by social inequalities, living and working conditions, and power dynamics. Therefore, developing technologically more sophisticated epidemiological models is not enough if they fail to incorporate the structural drivers that leave certain social groups more exposed, more vulnerable, or less equipped to protect themselves. At UPF, we will contribute our expertise in ecosocial determinants, health inequalities, and public policy, leading the design and evaluation of interventions that consider both efficacy and equity impacts. Our challenge is to ensure that AI goes beyond better predicting epidemics to helping explain them and developing socially just public health responses—built with stakeholder participation and capable of reducing, rather than reproducing, existing inequalities,” said Joan Benach, Director of the Health Inequalities Research Group – Employment Conditions Network (GREDS-EMCONET) and Co-Director of the JHU-UPF Public Policy Center at UPF.

In practical terms, AIBEM will treat equity as a primary model outcome rather than an afterthought. The age-based contact matrices that underpin most epidemic models will be expanded to include socioeconomic dimensions—occupation, income, and neighborhood—drawn from Spanish census data, labor force surveys, and stratified epidemiological records. This means disparities in disease burden and in the capacity to respond to an outbreak will emerge naturally from each simulated group’s material conditions, rather than being imposed by the modelers afterward. The approach echoes one of the hardest lessons of the COVID-19 pandemic: that a single national policy can produce profoundly unequal outcomes depending on who can work from home, who lives in crowded housing, and who can afford to isolate.

The project also embeds an ethical and governance framework aligned with the European AI Act, including bias audits, human oversight, and independent equity reviews. All models, datasets, and validation protocols will be published open-access following FAIR principles, making the methodology available for scrutiny and reuse. The expected deliverables include behavioral agents validated against real human experimental data with explicit socioeconomic diversity, coupled epidemiological models that quantify how behavioral heterogeneity modulates transmission, a systematic methodology for designing and evaluating interventions under realistic social conditions, and a decision-support toolkit for Spanish and European health authorities featuring policy briefs, ethical guidelines, and simulation software.

If the consortium succeeds, the implications could extend well beyond the next epidemic. Generative agent-based models are being explored across the social sciences as a way to simulate collective phenomena that resist traditional mathematical treatment, and AIBEM’s insistence on experimental validation against real participants offers a template for how such simulations might earn scientific credibility. For public health officials, the promise is more immediate: tools that can test, in silico, how a mask mandate, a vaccination campaign, or a lockdown will play out across communities with different levels of trust, income, and exposure—before those measures are imposed on real populations. The researchers caution that the challenge of capturing human nature remains formidable, but they argue that the arrival of generative AI has shifted the balance of that contest. The mathematics of pathogens, they note, has been solved; the mathematics of us is now within reach.

Subject of Research: Generative AI-driven behavioral epidemic modeling for public health intervention design

Article Title: UC3M participates in a pioneering scientific project using AI to anticipate human behavior during epidemics

Article References: UC3M participates in a pioneering scientific project using AI to anticipate human behavior during epidemics. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: AIBEM, epidemiological modeling, generative agents, large language models, agent-based models, human behavior, health inequalities, public health policy, UC3M, University of Zaragoza, behavioral experiments, la Caixa Foundation

The research team, led by Lei Lin of the Southwest University of Political Science and Law, together with Jing Tan of Chongqing University and Di Zheng of the Sichuan Academy of Social Sciences, recruited 599 urban residents in China for two randomized online experiments. The choice of population was deliberate. China has more than 900 million urban permanent residents, a group that overlaps heavily with the audiences of official government microblogs, weather applications, and other digital warning channels, and these users are among the most frequent adopters of AI-enabled services. The experimenters used Credamo, a survey platform with verified users, and applied strict quality controls: eligibility was restricted to the highest credit ratings, duplicate submissions from the same IP address or device were blocked, minimum completion times were enforced, and attention and comprehension checks filtered out careless respondents. Of 636 initial responses, 37 were excluded under preregistered criteria, leaving a final sample whose demographic profile shifted negligibly after cleaning.

The scenario at the heart of both experiments was a heavy rainfall warning issued during peak flood season in a Chinese city, a hazard chosen for its familiarity, its low political sensitivity, and the fact that both government agencies and AI platforms plausibly issue such alerts. Warning severity was communicated through the standardized color-coded signals of the China Meteorological Administration, ranging from blue for general risk through yellow and orange to red for extreme risk. All experimental materials were identical in content, length, and style; only the source cue varied. Pilot testing confirmed that participants could accurately identify who was said to have issued each warning and perceived the scenarios as realistic.

The first experiment employed a three-group between-subjects design comparing a government-only warning, an AI-only warning, and a consistent joint warning issued by both. Three hundred valid participants were randomly assigned, 101 to the government condition, 100 to the AI condition, and 99 to the joint condition, and balance checks confirmed no pretreatment differences across groups on gender, age, education, income, occupation, or disaster experience. A one-way analysis of variance revealed a substantial effect of warning source on information trust, with a partial eta squared of 0.215, a medium-to-large effect. Post hoc comparisons using Tukey’s honestly significant difference test showed that trust in government warnings significantly exceeded trust in AI warnings, with a mean difference of 0.29 points on the five-point scale and a Cohen’s d of roughly 0.45.

More striking still was the performance of the joint condition. Warnings attributed jointly to the meteorological station and an AI-driven warning platform produced the highest trust of all, exceeding the government-only condition by 0.26 points and the AI-only condition by 0.55 points, the latter corresponding to a Cohen’s d of approximately 0.85, a large effect. The authors interpret this as evidence that credibility emerges not from any single source but from the combination of institutional authority and technological channels. When AI alerts appear alongside official government communication, they are more likely to be perceived as complementary rather than autonomous, easing concerns about algorithmic opacity and unclear responsibility. The finding echoes the long-standing ‘speak with one voice’ principle in risk communication, which emphasizes coherent messaging across sources to sustain public confidence.

The experiment also confirmed that trust translates into action. In a hierarchical regression controlling for gender, age, education, prior disaster experience, AI usage, institutional trust, and risk perception, adding information trust to the model raised the explained variance in protective intentions from a nonsignificant 2.9 percent to 12.7 percent. Information trust was the strongest predictor of intentions to monitor hazard information, move vehicles and belongings to safety, identify emergency shelters, and stockpile supplies, with a standardized coefficient of 0.416. Risk perception retained a smaller but significant effect. Protective intention items showed acceptable internal consistency, with Cronbach’s alpha of 0.72, while the trust and anticipated regret composites reached 0.82 and 0.84 respectively.

The second experiment turned to the harder question: what happens when government and AI disagree? Four conditions were constructed. In one, both sources issued a low-level blue warning; in a second, both issued an orange warning, serving as a reference and replicating the joint condition from the first study. In the two conflict conditions, one source issued orange while the other issued blue, with the order of severity reversed between the two versions. After excluding invalid responses, 387 participants were distributed almost evenly across the four groups, and randomization checks again showed no meaningful imbalances. A pretest had verified that a one-level discrepancy was perceived as a genuine conflict and that orange was consistently read as more severe than blue.

Conflict, it turned out, produced a powerful emotional response. A one-way analysis of variance revealed a large effect of warning consistency on anticipated regret, with a partial eta squared of 0.349, and a nonparametric Kruskal-Wallis test corroborated the result. Participants in both conflict conditions reported significantly higher anticipated regret than those in either consistent condition, exceeding the consistent-low group by roughly 0.85 to 0.91 points and the consistent-high group by 0.33 to 0.39 points, all differences highly significant. Crucially, the two conflict groups did not differ from each other, meaning it did not matter whether the government or the AI issued the more severe alert. When sources diverged, people temporarily set aside questions of institutional authority and focused instead on avoiding the cost of underestimating the threat, a pattern consistent with worst-case thinking described in regret theory and loss aversion research.

Anticipated regret also proved to be a formidable driver of behavior. In a hierarchical regression, adding anticipated regret to the controls raised explained variance in protective intentions from a nonsignificant 1.7 percent to 26.8 percent, with a standardized coefficient of 0.502, meaning a one-unit increase in expected regret corresponded to roughly half a unit more protective intention, all else held constant. Together, the two experiments support a dual-pathway framework of decision making in multi-source warning environments: under consistent messaging, responses are structured by cognitive trust grounded in institutional authority and technological support; under conflicting messages, responses are instead guided by the emotional motivation to avoid future regret.

The implications for warning system design are substantial. Because joint government-AI releases enhance confidence, the authors suggest that policymakers develop coordination mechanisms between meteorological agencies and AI service providers, aligning update schedules, severity thresholds, and release timing to reduce discrepancies before they reach the public. At the same time, they caution against exploiting regret as a persuasive tool, which could breed anxiety or distrust; instead, communication strategies should explain why different systems produce different alerts and offer graded recommendations for how to respond. The study acknowledges limitations: the simulated scenarios cannot fully reproduce the urgency of real disasters, the sample of urban online users limits generalizability, and the two experiments examined trust and regret in separate randomized conditions rather than estimating a unified causal pathway. Nonetheless, as AI becomes an increasingly visible participant in public warning dissemination, the research makes clear that institutional credibility remains the anchor of public confidence, and that when machines and governments speak in harmony, people listen most closely of all.

Subject of Research: Public trust and behavioral responses to disaster warnings issued by governments versus artificial intelligence systems.

Article Title: Public Responses to Disaster Warnings from Government Versus AI: Evidence from Behavioral Experiments

Article References: Lin, L., Tan, J., & Zheng, D. (2026). Public Responses to Disaster Warnings from Government Versus AI: Evidence from Behavioral Experiments. International Journal of Disaster Risk Science. https://doi.org/10.1007/s13753-026-00766-2

Image Credits: AI Generated

DOI: 10.1007/s13753-026-00766-2

Keywords: disaster warnings, artificial intelligence, government trust, anticipated regret, risk communication, behavioral experiments, early warning systems, protective behavior, algorithm aversion, flood warnings, China, decision making