The latest session of the AI4PEP Lecture Series offered more than a discussion about artificial intelligence in health. It became a deeper reflection on what meaningful innovation should look like in a world still navigating the long-term impacts of COVID-19.
In his lecture, “Minimalism in Artificial Intelligence in Health after the COVID-19 Pandemic,” César Ugarte-Gilchallenged researchers, policymakers, health professionals, and innovators to reconsider the growing race toward increasingly complex AI systems. Instead of asking how sophisticated AI can become, he encouraged participants to ask a different question: What kind of AI is actually useful for the realities communities face?
The session explored a central idea that resonated strongly throughout the discussion: “AI is a tiny solution in an ocean of problems.” Rather than dismissing AI, the statement reframed its role. Dr. Ugarte-Gil emphasized that AI should support health systems, not overwhelm them with costly, disconnected, or unsustainable technologies.
As a physician-epidemiologist working at Universidad Peruana Cayetano Heredia, his perspective was grounded in real experiences implementing digital health tools in resource-constrained environments. Throughout the lecture, he shared examples from Peru and other global contexts to demonstrate how technologies that perform well in one setting can fail in another if they are not adapted to local realities.
One example focused on AI-driven cough analysis tools for respiratory disease detection. Models trained using data from Montreal did not perform reliably when tested in Lima, and vice versa. According to Dr. Ugarte-Gil, this highlights a critical challenge in AI development: machine learning systems are deeply shaped by the data used to train them. Without validation across multiple regions and populations, algorithms risk producing biased or ineffective results.
This issue of “generality” became one of the lecture’s key themes. AI systems developed in technologically advanced hospitals with modern equipment may not translate effectively into clinics with unstable internet access, limited infrastructure, or different cultural and environmental conditions. As he noted, an AI solution that works in Boston may not automatically work in Lima, Cape Town, or rural communities elsewhere in the Global South.
The lecture also raised concerns about interoperability and vendor lock-in. Governments, he explained, are increasingly pressured to choose between rapidly evolving AI tools, many of which cannot integrate with existing health information systems. Instead of simplifying care delivery, disconnected systems often create parallel workflows that burden already overstretched health workers.
This challenge becomes especially serious in low- and middle-income countries, where ministries of health must make long-term investment decisions under financial constraints. Dr. Ugarte-Gil described how governments are sometimes forced to choose technologies that may become obsolete within only a few years, while simultaneously managing the growing costs of subscriptions, upgrades, and proprietary systems.
Another major theme was ethics, governance, and privacy.
The lecture highlighted how AI-driven health systems increasingly rely on sensitive data such as biometric information, geolocation tracking, and wearable device monitoring. While these technologies can support disease surveillance and public health responses, they also raise urgent questions:
Who owns the data? How is it used? Who benefits from it? And who is protected when systems fail?
Dr. Ugarte-Gil warned that communities with low digital literacy are often the most vulnerable in these conversations. He stressed that digital inequality is not only a Global South issue, but a global one. Even among health professionals, levels of digital literacy vary significantly, affecting both adoption and trust in AI systems.
One of the lecture’s strongest contributions was its argument for “minimalism” in AI for health.
Rather than prioritizing complexity, Dr. Ugarte-Gil advocated for:
- Context-appropriate solutions
- Utility over sophistication
- Equity over speed
- Collaboration over competition
He proposed that AI systems should be:
- Explainable and auditable
- Open-source where possible
- Designed for long-term sustainability
- Capable of operating in low-bandwidth or offline environments
- Built with interoperability in mind
- Co-created with communities rather than imposed upon them
The session concluded with a broader reflection on the future of AI in global health. Participants emphasized that minimalism does not mean doing less innovation. Rather, it means designing technologies that are simpler, more adaptable, and more responsive to the realities of the communities they aim to serve.
In a rapidly evolving technological landscape, the lecture served as a reminder that impactful AI is not defined by complexity alone. Sometimes, the most effective innovation is the one that works reliably, ethically, and sustainably in the places where it is needed most.
The full lecture is now available on YouTube:
