A consortium led by the College de Zaragoza, with the participation of Universidad Carlos III de Madrid (UC3M), has been chosen within the 2026 Well being Analysis Name by the “la Caixa” Basis—probably the most aggressive biomedical analysis funding calls in Spain and Portugal. The AIBEM undertaking will develop synthetic intelligence (AI)-driven epidemiological fashions able to replicating how belief in well being authorities, adherence to tips, misinformation, socioeconomic inequalities, and social determinants affect the unfold of illness, whereas concurrently evaluating the effectiveness of public well being measures to mitigate group transmission.
For 20 years, science has allowed us to precisely mannequin the dynamics of pathogen unfold by way of contact networks. Nonetheless, the true problem has all the time been the unpredictability of our personal nature. Capturing the complexity of human habits throughout a well being disaster stays an unresolved problem. The rise of generative AI adjustments the foundations of the sport in our favor: by equipping autonomous computational brokers with reasoning capabilities and mutual affect, we will remodel epidemiological fashions from mere mathematical projections right into a trustworthy reflection of the human issue on a societal scale.”
Yamir Moreno, undertaking’s principal investigator, Director of the Institute for Biocomputation and Physics of Advanced Programs (BIFI) and member of the Division of Theoretical Physics, College de Zaragoza
“UC3M’s foremost contribution is its intensive expertise in designing and executing behavioral experiments with human members,” notes Anxo Sánchez, one other researcher on the undertaking from UC3M’s Interdisciplinary Group of Advanced Programs (GISC). “This functionality is vital, as validating generative brokers requires not solely evaluating their habits in opposition to commonplace experiments, but additionally modeling epidemic-related situations and designing experiments that faithfully replicate them—a lot of which haven’t been addressed earlier than and will likely be essential to the undertaking’s success,” he concludes.
Fashions geared toward emulating human habits
The COVID-19 pandemic demonstrated that compliance with measures reminiscent of vaccination, mask-wearing, or mobility restrictions different considerably throughout social teams and over time, pushed by danger notion, belief in establishments, misinformation, socioeconomic atmosphere, vulnerability, and materials inequalities. However, commonplace epidemiological fashions proceed to deal with human habits as a set or exogenous parameter, considerably limiting their utility as decision-making instruments.
On this regard, AIBEM proposes a paradigm shift: generative agent-based fashions (GABMs), the place every simulated particular person is powered by a big language mannequin (LLM) configured with a selected profile. This is identical underlying know-how powering AI platforms like ChatGPT or Gemini. These brokers will possess their very own identification, targets, reminiscence, and instruments. Inside this framework, they’ll work together with each other, make choices, set expectations, and generate social norms. Moreover, their responses will likely be systematically validated in opposition to experiments involving actual human members.
“Epidemics usually are not purely organic phenomena: their unfold and penalties are deeply conditioned by social inequalities, residing and dealing situations, and energy dynamics. Due to this fact, creating technologically extra subtle epidemiological fashions shouldn’t be sufficient in the event that they fail to include the structural drivers that depart sure social teams extra uncovered, extra susceptible, or much less geared up to guard themselves. At UPF, we’ll contribute our experience in ecosocial determinants, well being inequalities, and public coverage, main the design and analysis of interventions that think about each efficacy and fairness impacts. Our problem is to make sure that AI goes past higher predicting epidemics to serving to clarify them and creating socially simply public well being responses—constructed with stakeholder participation and able to lowering, somewhat than reproducing, present inequalities,” notes one other researcher on the undertaking, Joan Benach, Director of the Well being Inequalities Analysis Group – Employment Circumstances Community (GREDS-EMCONET) and Co-Director of the JHU-UPF Public Coverage Heart at Universitat Pompeu Fabra (UPF).
The undertaking is structured into three foremost work packages. UC3M will conduct behavioral experiments with socioeconomically various members utilizing the IBSEN platform, developed inside the framework of the European H2020 undertaking of the identical title. In the meantime, the College of Zaragoza will develop generative brokers that replicate human habits and combine them into epidemiological fashions, leveraging BIFI’s supercomputing infrastructure (CESAR) and the intensive experience of the Advanced Networks and Programs Group (COSNET) in mathematical and computational epidemic modeling. UPF will translate the outcomes into decision-support instruments for well being authorities, together with an express evaluation of the distributional affect of every intervention.
Fairness as a mannequin final result somewhat than an afterthought
One in every of AIBEM’s defining options is that fairness shouldn’t be appended on the finish of the evaluation, however handled as a main final result of the mannequin itself. Age-based contact matrices will likely be expanded to incorporate socioeconomic dimensions—occupation, earnings, neighborhood—drawn from Spanish census information, labor power surveys, and stratified epidemiological data, making certain that disparities in illness burden and response capability emerge naturally from every group’s materials situations.
The undertaking additionally incorporates an moral and governance framework aligned with the European AI Act (EU AI Act), that includes bias audits, human oversight, and impartial fairness critiques. All fashions, datasets, and validation protocols will likely be revealed open-access following FAIR ideas.
Anticipated deliverables embrace behavioral brokers validated in opposition to actual human experimental information with express socioeconomic variety, coupled epidemiological fashions quantifying how behavioral heterogeneity modulates transmission, a scientific methodology to design and consider interventions below real looking social situations, and a decision-support toolkit for Spanish and European well being authorities that includes coverage briefs, moral tips, and simulation software program.
The undertaking, titled AI-Pushed Behavioural Epidemic Fashions for Intervention Design (AIBEM), will obtain €970,139 over three years. AIBEM was considered one of 32 tasks chosen out of 800 proposals submitted to the 2026 name, representing a hit charge of 4%. In whole, the “la Caixa” Basis will award €25.7 million to the tasks chosen on this version.
Supply:
Universidad Carlos III de Madrid (UC3M)
