Using AI To Detect the Next Zoonotic Threat
AI may help us see the next zoonotic threat coming—but are surveillance systems ready?
It is estimated that over 60% of all known infectious diseases and 75% of new, emerging diseases in humans originate from animals. The COVID-19 pandemic and the recent avian influenza and Ebola outbreaks highlight the continual risk for zoonotic virus spillover anywhere humans, animals, and the environment intersect. As climate change, habitat loss, urbanization, and global travel increase opportunities for pathogen transmission across species, preventing future epidemics or pandemics requires surveillance systems capable of detecting threats earlier than ever before.
Artificial intelligence (AI) has emerged as a powerful tool for strengthening surveillance by rapidly analyzing datasets that are too large and/or complex for traditional surveillance methods to process efficiently. Unlike conventional systems that rely on manual reporting, AI can rapidly analyze vast amounts of information from veterinary records, environmental monitoring, climate data, genomic sequencing, electronic health records, and patterns of animal and human movement to detect potential spillover events before they can develop into widespread outbreaks.
For known zoonotic pathogens, AI predictive models can identify countries and/or regions at greatest risk of disease transmission by integrating geospatial, climate, ecologic, demographic, and mobility data.
Rather than reacting to outbreaks as disease cases spike, health officials can initiate targeted testing and vaccination (if available), implement control measures to limit disease transmission, and work with local hospitals to ensure sufficient staffing and resources. In addition, AI-assisted analysis of viral genome sequences can identify highly transmissible viral variants and/or mutations that may reduce the effectiveness of currently available vaccines or antivirals.
Identifying zoonotic threats earlier
AI cannot prevent the next zoonotic disease outbreak in isolation. Its success depends on equitable data, thoughtful implementation, and a generation of health professionals equipped to use it responsibly.
Many AI models are trained using information culled from high-income countries and large urban medical centers since these settings possess the digital infrastructure and extensive datasets ideal for AI model development. However, zoonotic diseases typically emerge in rural regions and low- and middle-income countries where resources for disease surveillance and digital infrastructure are limited. Datasets from these regions are often incomplete or fragmented, and underserved populations are often underrepresented. This imbalance creates a significant risk for algorithmic bias.
An AI model trained primarily on data from urban hospitals in North America or Europe may perform poorly when applied to regions where endemic diseases are more prevalent and ecology, environmental conditions, and access to healthcare may differ substantially. These models may fail to recognize early signs of a spillover, underestimate transmission rates and the risk of an outbreak, or misdirect the allocation of vital supplies (e.g., testing kits, personal protective equipment, vaccines).
For example, early AI diagnostic models for COVID-19 trained on digital chest X-ray images from large academic hospitals in high-income countries were found to perform worse in low-income countries due to differences in imaging equipment, clinical workflow, and patient populations. Similarly, algorithms designed to map the spread of COVID-19 that relied on smartphone usage excluded low-income, older, and rural populations lacking digital access from disease protection efforts. Models to predict avian influenza outbreaks developed using surveillance data from large commercial farms in North America and Europe were found not to translate well in low-income countries where smallholder farms predominate.
Lacking representative data, models with algorithmic biases could unintentionally reinforce global health inequities by directing surveillance efforts and resources towards regions already well represented in the data and away from regions and communities that need them most. Such an outcome can erode trust in AI and strain relationships between stakeholders in zoonotic disease prevention (i.e., local communities, medical professionals, wildlife experts, government health authorities, non-governmental organizations).
Making AI surveillance more equitable
Bridging the data gap that exists in low- and middle-income countries to increase AI inclusivity requires aggressive investment in local digital and surveillance infrastructures and robust international collaborations focused on the development of datasets that are more diverse and representative of the demographics of disease-risk regions. In addition, the inclusion of local experts in AI validation and post-deployment monitoring is essential for mitigating bias and ensuring safety and accuracy across demographic groups.Notable initiatives to improve AI infrastructure in emerging economies to strengthen disease surveillance and outbreak management include the Artificial Intelligence for Pandemic and Epidemic Preparedness and Response Network, WHO Regional Office for Africa’s Preparedness Data Exchange, and the Horizon 1000 program.
At the same time, successful integration of AI in disease surveillance and prevention requires healthcare and public health professionals who understand AI’s capabilities and limitations. Likewise, informed human oversight is essential for ensuring the responsible use of AI in healthcare. Accordingly, preparing an AI-literate healthcare workforce is a priority. Medical, nursing, pharmacy, and public health curricula should incorporate training on machine learning models, interpretation of AI outputs, algorithmic bias, data security, and the legal and ethical issues surrounding AI use in healthcare.
Several groups focused on professional education (e.g., Association of American Medical Colleges, International Pharmaceutical Federation) have already begun developing competencies for AI and digital health.
Why AI literacy matters in disease surveillance
Providing students with hands-on experiences analyzing complex datasets and evaluating predictive models will help break down the “black box” nature of AI models, cultivate critical thinking and creativity, and foster proficiency in deploying AI technologies to address real-world health challenges. As more “AI-literate” graduates enter the workforce, they can be paired with experienced clinicians, researchers, and public health practitioners who have extensive real-world experience and an intuitive understanding of disease patterns but little formal education in AI.Coupling the complementary strengths of early-career and established professionals will help ensure teams are better equipped to navigate the uncertainty of emerging or mutating zoonotic threats. Newer grads are more adept at critically evaluating AI systems, interpreting model outputs, understanding how training data influences model predictions, and recognizing when algorithmic bias may affect model recommendations.
Experienced professionals are more adept at interpreting ambiguous clinical findings, navigating complex health data privacy regulations, and recognizing when an AI-generated recommendation conflicts with the realities of patient care or public health practice.
Formal and informal knowledge sharing can benefit both generations. Experienced professionals can mentor early-career researchers in clinical reasoning and data governance, while early-career researchers can share expertise in AI and digital health.
This exchange can strengthen model validation and AI oversight and drive innovation in detecting and responding to viral zoonoses.
Final thoughts
As new viruses continue to emerge at the human‒animal‒environment interface, AI has the potential to revolutionize how we detect and prevent viral zoonoses. However, AI will only reach its full potential if the systems we build are as equitable as they are innovative.
Achieving this will require researchers and policymakers to prioritize model transparency, rigorous external validation across diverse populations and geographic regions, and continuous monitoring for bias. At the same time, AI literacy must become a core competency for both current and future healthcare professionals, empowering them to question, interpret, and apply AI outputs responsibly.
Add Technology Networks as a preferred Google source to see more of our trusted coverage.
