In this guest contribution, Frank Tu Ngo at the Global AMR R&D Hub discusses the role of artificial intelligence in curbing antimicrobial resistance (AMR). He introduces an innovative tool that automates the mapping of funding data through machine learning, highlighting both its advantages and limitations in addressing this critical issue.
Antimicrobial resistance (AMR) has been called a “silent pandemic” by the World Health Organization. While COVID-19 claimed over 7 million lives in four years, AMR is projected to cause up to 10 million deaths annually by 2050.
The economic toll is just as staggering. COVID-19 cost the global economy an estimated $12.5 trillion. AMR, if unchecked, is expected to reduce global GDP by $3.4 trillion every year by 2030. That’s equivalent to giving 10 million students a full scholarship to a top U.S. university — every year.
Addressing AMR requires coordinated global action: better policies, strong regulations, and smarter funding. Yet one of the biggest challenges is fragmentation. Funding streams, innovation pipelines, and evidence are often siloed and outdated. What’s urgently needed is a shared, transparent platform that brings this information together to guide more effective responses.
That’s why, in 2017 during the German G20 Presidency, world leaders called for a new international organisation to support collaboration on AMR research and development. The result was the creation of the Global AMR R&D Hub, where I currently work. The Hub is governed by 19 countries, the European Commission, the Gates Foundation, and the Wellcome Trust, and works closely with five observers: the Africa CDC, WHO, FAO, WOAH, and the OECD.
From manual tracking to machine learning
At the core of our work is the Dynamic Dashboard — an online tool that maps AMR-related funding from over 100 countries. It helps governments, researchers, and funders understand where resources are going and how to better target their efforts.
But updating the Dashboard has been a laborious task. To date, our small team has manually collected and categorised over 17,000 projects across dozens of categories and subcategories. We aim to process around 500 new projects monthly, but manual workflows have resulted in a lag of 1–2 years in fully capturing each financial year.
To overcome this, we launched an AI pilot project in 2025 to automate the classification of AMR funding data using machine learning. With the AI model in place, we can now process 1,000–2,000 funding entries per month — a threefold increase in speed. The model has already achieved 80–90% accuracy in several categories, with human experts still overseeing quality control.
Informing smarter policies, faster
What we’ve learned from implementing this AI pilot is that the innovation has the potential to reshape how AMR policies are made. With better data, countries can allocate funding more effectively, reduce duplication, and identify neglected areas.
For example, Canada used our data to support a national antibiotic access initiative. Other countries have used the Dashboard to shape funding calls and align national strategies with global efforts. Faster data means faster decisions — and more timely action.
In low- and middle-income countries, where capacity for data analysis is often limited, centralised platforms like ours can be particularly valuable. By providing real-time, validated intelligence, we can help reduce the barriers to informed policy participation and coordination.
Barriers on the road to innovation
Still, unlocking the full potential of AI in AMR tracking comes with real-world challenges — both technical and policy-related.
Technically, building these systems requires a combination of advanced programming skills and in-depth domain knowledge. When we launched our pilot, this became immediately clear: our programmer wasn’t an AMR expert, and our AMR team wasn’t trained in coding. It took time — and mutual learning — for everyone to align. This disconnect is likely common in digital health projects and highlights the need for stronger collaboration between technical and subject-matter experts from the outset.
AI models themselves also have limitations. While they’re evolving rapidly, current systems can still struggle with nuance and context. Our own pilot encountered this, with some categories performing much better than others. We remain hopeful that emerging technologies will address these limitations in future iterations.
On the policy side, fragmentation continues to be a hurdle. AMR funding is already dispersed and limited, and introducing new digital tools — particularly unproven ones — can seem risky to funders and institutions. Securing long-term support for development, integration, and maintenance is not guaranteed.
That said, there are promising ways forward. For one, adapting open-source AI models from different fields could help AMR efforts leapfrog some early development hurdles. As we learned through conversations with data science specialists and policymakers, relevant solutions may already exist in other domains — they just need tailoring to AMR. This approach could accelerate progress while avoiding the trap of reinventing the wheel.
A model for other global challenges
While our immediate focus is AMR, the broader implications of this work are increasingly evident. The same AI-driven methods we’ve piloted can be applied to other complex and fragmented policy areas — from tracking climate finance, mapping education investments, to tracking digital health or future pandemic preparedness funding. In all these domains, vast amounts of data already exist. What’s missing is structure, clarity, and accessibility. Machine learning offers a path forward — transforming data into decision-ready intelligence.
If we position digital AMR platforms not just as specialised tech, but as public-interest digital infrastructure with broad utility, their value becomes clearer. With the right framing and partnerships, these tools can move from niche innovation to essential systems support.
By designing and sharing these tools openly, we hope to spark a shift toward transparent data systems that can support more coordinated action — across countries, sectors, and crises.
Ultimately, this isn’t just about speeding up data classification or improving our dashboard. It’s about equipping global health and policy communities with the infrastructure they need to respond faster, more effectively, and more collaboratively — to the challenges we face now and the ones still to come.
Photo by Michael Schiffer on Unsplash.
