ESCAIDE 2025: AI in Public Health
At this year’s ESCAIDE conference in Warsaw, the session “Artificial intelligence in public health: hype or help?” brought together experts from Europe and the U.S. to discuss the potential and challenges of applying artificial intelligence in public health. Moderator Laura Espinoza from the ECDC opened the session by emphasizing its goal: participants should leave with an understanding of how AI can be applied in practice, not just considered theoretically.
Espinoza explained that AI is not just about large language models, but a broader spectrum of technologies that enable machines to recognize patterns and process natural language. Key challenges include the gap between rapid innovation and the slower development of AI literacy in public health, the need for fair and open approaches, and regulation and accountability for AI-driven decisions.
The panelists were Dr. Sergio Consoli from the European Commission’s Joint Research Centre, Professor Yanis Paschalidis from Boston University, and Annie Hartley from the LIGHTE laboratory.
How AI Is Changing Public Health
The first speaker, Dr. Sergio Consoli from the European Commission’s Joint Research Centre, presented digital health tools for public health. His team uses the Epidemic Intelligence from Open Source (EIOS) platform to collect epidemiological reports from around the world. Faced with a massive volume of data, Consoli and his colleagues developed a combined approach that merges commercial and open-source language models, allowing precise extraction of key outbreak information. To resolve differences across reports, they use ontologies and knowledge graphs, creating an epidemiological knowledge graph that is updated daily. Consoli’s team also experiments with methods such as prompt engineering and retrieval-augmented generation to analyze and generate relevant real-time data.
Professor Yannis Paschalidis from Boston University presented BEACON, the first global open-source system for real-time outbreak surveillance. The system uses an LLM called Pandemic Llama to automatically extract key epidemiological data, while experts review and verify the accuracy of the reports. Professor Paschalidis emphasized that smaller models trained on high-quality data often outperform larger, less reliable models, highlighting the importance of data sources and quality.
Annie Hartley from the LIGHTE laboratory presented her team’s approach to separating hype from real utility in AI. Her team develops open models such as Apertus and platforms like Meditron and MOVE, which enable continuous evaluation and adaptation of models to local contexts. Hartley stressed that AI is genuinely valuable in public health only when it is grounded in evidence and expert evaluation, rather than marketing claims and promises.
Collaboration and resource sharing are crucial
The panel discussed challenges such as access to data and computational resources, particularly in academia. Collaboration and resource sharing are crucial to enabling AI training and application for public good rather than commercial purposes.
In closing, the speakers offered advice to public health agencies: Consoli emphasized the importance of robust infrastructure and ongoing learning; Paschalidis highlighted interdisciplinary collaboration; and Hartley recommended continuous validation and evidence generation to avoid hype and ensure real benefit to communities.
Image: ESCAIDE 2025
