Pr. Jude Kong is a Canada Research Chair in Community-Oriented AI and Mathematical Modeling at the University of Toronto and member of the College of the Royal Society of Canada (RSC), pioneers equitable, mathematical and AI-driven solutions for global health crises. He is the founder and director of the AI4PEP network, which comprises over 210 researchers from 21 countries, focusing on leveraging Southern-led, responsible AI solutions to enhance public health systems for more effective prevention, preparedness, and response to disease outbreaks.
1/ Can you give us examples of how you use environmental data to improve health prevention in the projects you support?
Across the projects we support, environmental data functions as a critical early warning layer that enables a shift from reactive health responses to preventive, anticipatory decision-making. By integrating environmental signals with epidemiological and behavioural data, our AI-powered One Health platforms can detect conditions that precede disease outbreaks (often weeks in advance) allowing ministries of health and communities to intervene early.
At the core of these systems is a comprehensive environmental data stack which includes air quality indicators, meteorological variables, data on wastewater-based pathogen and antimicrobial resistance (AMR), on pollution, and so many more.
Together, these data streams enable a unified environmental intelligence layer that feeds into AI-driven early warning systems tailored to each country’s epidemiological context.
Country-level examples include:
- Indonesia:Environmental data augments the national EWARS system across multiple disease categories. For example, flooding, rainfall, and wastewater indicators inform waterborne disease risk (e.g., cholera), while air pollution data supports forecasting of respiratory diseases such as COVID-19 and TB. Mobility data further refines transmission modelling.
- Peru:Drone-based environmental surveillance enables high-resolution mapping of mosquito habitats, including water containers and deforestation boundaries. Combined with climate and acoustic data, this supports precise species classification and localized outbreak risk prediction.
- Dominican Republic:An eco-epidemiological early warning system integrates climate variables, satellite and drone data, and entomological indicators to predict dengue, Zika, and chikungunya outbreaks, with climate change projections embedded into the models.
- Philippines:Climate and disaster-related environmental indicators (rainfall, flooding) are linked with telehealth data to predict both infectious disease risk and secondary impacts such as mental health and displacement.
- Ethiopia, Burundi, DRC, Nigeria, and Jamaica:Across these countries, environmental data layers (air quality, climate, land use, and wastewater) are integrated into national early warning systems targeting zoonotic, vector-borne, waterborne, and respiratory diseases, tailored to each country’s risk profile.
2/ What are the current challenges in terms of One Health data governance that prevent its full exploitation, for example through AI tools?
The honest answer is that One Health data governance does not yet exist in any meaningful operational sense. We have the concept and we have some inspiring national and regional initiatives. But when you actually try to build an AI system that integrates human, animal, and environmental data across sectors and across borders (as AI4PEP has done in 23 countries) you encounter a series of structural failures that no amount of political goodwill has yet resolved.
The most fundamental challenge is fragmentation. Health, animal, and environmental data are generated and held by entirely separate institutional systems with almost no systematic integration. The result is predictable: AI models trained on any single data stream miss the very interactions that give One Health its predictive power.
Another key challenge is that One Health data governance focuses too much on access, not enough on balancing it with trust. Many organisations remain understandably cautious about sharing sensitive, operationally important, or politically consequential data (health records, agricultural biosecurity information, etc.). Governance frameworks that focus only on opening data without addressing the trust conditions under which institutions will actually share it are frameworks that do not work in practice.
Closely related is the quality and representativeness problem. One Health data is multimodal, unevenly collected, context-dependent, and often poorly documented. AI systems do not neutralise these weaknesses; they amplify them. A model trained on sparse, biased, or poorly provenanced data will produce confident predictions that reflect those biases at scale. Similarly, the communities that generate the most epidemiologically critical One Health observations are precisely the communities least represented in the datasets that AI systems learn from.
The third challenge is access to privately held data. Much of the most valuable data for One Health surveillance (mobile phone metadata for mobility modelling, social media signals for behavioural surveillance, purchasing patterns for food-system monitoring) is controlled by corporations whose primary incentive is not public health.
Governance frameworks themselves present a fourth challenge. Existing ethical review and data governance frameworks were designed for clinical trials with individual human subjects but rarely focus on data sovereignty, particularly for Indigenous communities. In our work with Amazonian Indigenous communities in Brazil, with the Baka, Bakola, and Bedzang forest peoples in Cameroon, and with Indigenous communities in Canada, we have repeatedly encountered situations where no existing framework provided a clear pathway for communities to own, control, or benefit from data generated about their health and environment.
There is also a structural capacity gap that we cannot afford to ignore. A small number of high-income countries dominate AI development in health, benefiting from advanced computing infrastructure, large and well-curated training datasets, and deep institutional expertise. AI systems designed in high-income settings and applied without adaptation to very different epidemiological contexts carry their designers’ blind spots with them.
Then there is the funding problem. Most One Health AI initiatives are funded as fixed-term projects, which is structurally incompatible with longitudinal surveillance systems, which need years of data to detect trends, train robust models, and validate predictions across multiple outbreak cycles. Pandemic preparedness that depends on two-year project funding is not preparedness. It is a repeated rehearsal for the next failure.
Finally, and this is something the field discusses less than it should: misinformation is not only a problem in public discourse, but contaminates the data streams themselves. Incomplete surveillance reports, inconsistent disease classification, self-reporting biases, and politically motivated underreporting are pervasive in the countries where we work.
Underlying all of these governance failures is a technical problem: there is no shared, privacy-preserving infrastructure that allows countries to collaborate on One Health AI models without surrendering sovereignty over their national data.
3/ The One Health Summit, to be held on 7 April in Lyon, will bring together the scientific community, civil society, private actors and heads of state. What solutions do you hope to see emerge from this event at the international or regional level?
When a summit like Lyon brings together scientists, civil society, private actors, and heads of state in the same room, the risk is that it produces a communiqué full of principles that everyone endorses and no one implements. The One Health concept has never lacked for inspiring statements. What it lacks is operational infrastructure – the actual data systems, governance frameworks, and technical platforms that would make it function in practice.
Concrete commitments should include:
- A globally accessible One Health data portal compliant with the FAIR principles (Findable, Accessible, Interoperable, Reusable) and the creation of regional One Health observatories
- A One Health social data observatory able to detect early signals of emerging threats through human digital behaviour
- Binding commitments to data sovereignty as an operational standard. This should include the recognition Indigenous communities’ rights
- Investment in capacity, health literacy, and community engagement to enable genuine co-creation, with communities as partners in surveillance design from the outset
- Catalyse investment in point-of-need diagnostic and response tools that are structurally integrated into One Health surveillance systems from the outset
- Integrate environmental, animal health, and community-level data into existing national surveillance platforms
- Establish a global federated learning platform for one health data to derive shared intelligence from those national systems across borders; without requiring any country to surrender sovereignty over its data.
What the One Health community needs from Lyon is not a declaration; it is a roadmap: one that supports shared data standards and governance frameworks, enables responsible AI adoption with equity and explainability at its centre, seeds regional implementation through observatories and federated pilot structures, and establishes realistic, measurable priorities that countries and institutions can adapt to their own realities and report against over time.
Scientific knowledge, political commitment, public trust, regulatory clarity, and local capacity must be aligned around the same objectives; and that alignment requires more than inspiring language. It requires the kind of sustained, accountable, multisectoral coordination that only concrete institutional commitments can deliver. The One Health concept has waited long enough for its data infrastructure to catch up with its ambitions. The tools exist. The evidence base exists. The communities are ready to partner. What is needed now is the political will to build systems that are global in , equitable in design, and genuinely accountable to the people they are meant to serve.
;
