America is Betting on AI While Ignoring Its Biggest Healthcare Weakness

The following is a guest article by Michael Savas, Fellow at the Center for Health Policy and Media Engagement at The George Washington University, and Richard Ricciardi, Professor and Executive Director, Center for Health Policy and Media Engagement at The George Washington University

Among the Trump administration’s recent executive orders is one with an ambitious title: Unlocking Cures for Pediatric Cancer with Artificial Intelligence. It directs federal agencies to deploy AI tools to transform how childhood cancers are diagnosed and treated. On paper, it signals a welcome embrace of evolving technologies and cancer care.

For families like that of Elena, a six-year-old recently diagnosed with leukemia, the promise is far more personal. AI could mean faster diagnoses, fewer failed treatments, and better odds of survival. But unless we confront a fundamental barrier in the U.S. health care system, how we manage and learn from data, AI’s promise risks colliding with a fragmented infrastructure that has stymied innovation for years.

For nearly two decades, vast amounts of patient data have been locked inside disconnected health systems, electronic health records, and research silos. As primary care scholar Kurt Stange has argued, this fragmentation is critically underappreciated, fueling inefficiency, poorer outcomes, and widening inequities, while inhibiting collaboration across specialties. Much of the collision traces back to well-intentioned policies like the 2009 HITECH Act, which accelerated adoption of electronic health records but failed to require meaningful interoperability. The result: systems optimized for billing and market share, not learning.

The consequences extend beyond provider frustration. They weaken America’s ability to innovate at scale, especially when compared with countries that treat health data as a public asset rather than a proprietary commodity. Artificial intelligence depends on large volumes of high-quality, diverse data to detect patterns and improve predictions, a particularly powerful proposition in healthcare. Countries with nationalized or centrally coordinated health systems, including Finland and the United Kingdom, enjoy a structural advantage: longitudinal, population-level datasets that can fuel continuous learning. These systems are not without flaws, but they are far better positioned to harness AI than the United States, which has barely begun laying the groundwork.

Yet centralized data collection in America triggers deep and persistent concerns about privacy, surveillance, and cybersecurity. In a 2019 survey by the U.S. Chamber of Commerce and Morning Consult, fewer than one in three Americans supported even a limited national claims database to improve price transparency, and nearly three-quarters expressed privacy concerns. Any strategy for AI-enabled health care must reconcile these anxieties with the reality that fragmented data undermines both innovation and care.

There is a path forward, and it does not require building a national data warehouse.

Federated learning, first proposed by Google researchers in 2016, allows AI models to be trained where data already reside, within hospitals, clinics, and research institutions, rather than moving sensitive patient records to a central repository. Each participating site trains the model locally and shares only what the algorithm learns, not the underlying data. Over repeated cycles, the model improves by drawing insight from millions of patient encounters, while patient information remains protected behind institutional firewalls.

At the national scale, federated learning could fundamentally reshape American healthcare. By continuously pooling insights across care settings, such a network could improve diagnostic accuracy, accelerate treatment discovery, identify emerging risks earlier, and help address persistent inequities, all without compromising patient confidentiality. Early implementations, including platforms developed at Johns Hopkins and Mayo Clinic, already demonstrate how this approach can advance both clinical care and research.

For Elena, this kind of system would mean her care is no longer informed only by the experience of a single institution, but by patterns drawn from thousands of similar cases nationwide. AI could help anticipate how she might respond to treatment, flag early warning signs, predict side effects, and identify regimens that have worked best for children like her. Her journey would contribute to a continuous learning system, one that improves outcomes not only for her but for the next child who has the same devastating diagnosis.

Skeptics are right to caution that no technology can, on its own, overcome the structural weaknesses of American healthcare. Federated systems must guard against sophisticated privacy breaches. Under-resourced hospitals could struggle to participate. Poor interoperability standards risk feeding low-quality data into powerful algorithms, and opaque models could undermine clinical judgment or introduce new legal risks.

These concerns are real, but increasingly addressable. Advances such as secure aggregation and differential privacy strengthen protection against data leakage. Flexible participation models lower barriers for rural and safety-net hospitals. Improved data governance and transparency tools help clinicians understand, trust, and appropriately use AI-generated insights.

What remains missing is sustained institutional commitment. Scaling federated learning will require deliberate public-private collaboration, potentially anchored within the Department of Health and Human Services, that brings together NIH scientific expertise, health systems, and responsible technology developers. With the right governance, such a partnership could transform today’s fragmented data landscape into a national learning ecosystem.

Without modernizing how the health care system learns from its own data, the United States risks squandering one of the most consequential opportunities in modern healthcare. Federated learning is not just a technical solution; it is a bridge from fragmentation to collective insight. If we fail to build it, families like Elena’s will continue to bear the cost of our inaction.

They deserve more than promises. They deserve a system that learns and gets better, for every child.

About Michael Savas

Michael Savas is a fellow at the Center for Health Policy and Media Engagement at The George Washington University.

About Richard Ricciardi

Richard Ricciardi is a Professor and Executive Director, Center for Health Policy and Media Engagement at The George Washington University.

The following is a guest article by Michael Savas, Fellow at the Center for Health Policy and Media Engagement at The George Washington University, and Richard Ricciardi, Professor and Executive Director, Center for Health Policy and Media Engagement at The George Washington University

Among the Trump administration’s recent executive orders is one with an ambitious title: Unlocking Cures for Pediatric Cancer with Artificial Intelligence. It directs federal agencies to deploy AI tools to transform how childhood cancers are diagnosed and treated. On paper, it signals a welcome embrace of evolving technologies and cancer care.

For families like that of Elena, a six-year-old recently diagnosed with leukemia, the promise is far more personal. AI could mean faster diagnoses, fewer failed treatments, and better odds of survival. But unless we confront a fundamental barrier in the U.S. health care system, how we manage and learn from data, AI’s promise risks colliding with a fragmented infrastructure that has stymied innovation for years.

For nearly two decades, vast amounts of patient data have been locked inside disconnected health systems, electronic health records, and research silos. As primary care scholar Kurt Stange has argued, this fragmentation is critically underappreciated, fueling inefficiency, poorer outcomes, and widening inequities, while inhibiting collaboration across specialties. Much of the collision traces back to well-intentioned policies like the 2009 HITECH Act, which accelerated adoption of electronic health records but failed to require meaningful interoperability. The result: systems optimized for billing and market share, not learning.

The consequences extend beyond provider frustration. They weaken America’s ability to innovate at scale, especially when compared with countries that treat health data as a public asset rather than a proprietary commodity. Artificial intelligence depends on large volumes of high-quality, diverse data to detect patterns and improve predictions, a particularly powerful proposition in healthcare. Countries with nationalized or centrally coordinated health systems, including Finland and the United Kingdom, enjoy a structural advantage: longitudinal, population-level datasets that can fuel continuous learning. These systems are not without flaws, but they are far better positioned to harness AI than the United States, which has barely begun laying the groundwork.

Yet centralized data collection in America triggers deep and persistent concerns about privacy, surveillance, and cybersecurity. In a 2019 survey by the U.S. Chamber of Commerce and Morning Consult, fewer than one in three Americans supported even a limited national claims database to improve price transparency, and nearly three-quarters expressed privacy concerns. Any strategy for AI-enabled health care must reconcile these anxieties with the reality that fragmented data undermines both innovation and care.

There is a path forward, and it does not require building a national data warehouse.

Federated learning, first proposed by Google researchers in 2016, allows AI models to be trained where data already reside, within hospitals, clinics, and research institutions, rather than moving sensitive patient records to a central repository. Each participating site trains the model locally and shares only what the algorithm learns, not the underlying data. Over repeated cycles, the model improves by drawing insight from millions of patient encounters, while patient information remains protected behind institutional firewalls.

At the national scale, federated learning could fundamentally reshape American healthcare. By continuously pooling insights across care settings, such a network could improve diagnostic accuracy, accelerate treatment discovery, identify emerging risks earlier, and help address persistent inequities, all without compromising patient confidentiality. Early implementations, including platforms developed at Johns Hopkins and Mayo Clinic, already demonstrate how this approach can advance both clinical care and research.

For Elena, this kind of system would mean her care is no longer informed only by the experience of a single institution, but by patterns drawn from thousands of similar cases nationwide. AI could help anticipate how she might respond to treatment, flag early warning signs, predict side effects, and identify regimens that have worked best for children like her. Her journey would contribute to a continuous learning system, one that improves outcomes not only for her but for the next child who has the same devastating diagnosis.

Skeptics are right to caution that no technology can, on its own, overcome the structural weaknesses of American healthcare. Federated systems must guard against sophisticated privacy breaches. Under-resourced hospitals could struggle to participate. Poor interoperability standards risk feeding low-quality data into powerful algorithms, and opaque models could undermine clinical judgment or introduce new legal risks.

These concerns are real, but increasingly addressable. Advances such as secure aggregation and differential privacy strengthen protection against data leakage. Flexible participation models lower barriers for rural and safety-net hospitals. Improved data governance and transparency tools help clinicians understand, trust, and appropriately use AI-generated insights.

What remains missing is sustained institutional commitment. Scaling federated learning will require deliberate public-private collaboration, potentially anchored within the Department of Health and Human Services, that brings together NIH scientific expertise, health systems, and responsible technology developers. With the right governance, such a partnership could transform today’s fragmented data landscape into a national learning ecosystem.

Without modernizing how the health care system learns from its own data, the United States risks squandering one of the most consequential opportunities in modern healthcare. Federated learning is not just a technical solution; it is a bridge from fragmentation to collective insight. If we fail to build it, families like Elena’s will continue to bear the cost of our inaction.

They deserve more than promises. They deserve a system that learns and gets better, for every child.

About Michael Savas

Michael Savas is a fellow at the Center for Health Policy and Media Engagement at The George Washington University.

About Richard Ricciardi

Richard Ricciardi is a Professor and Executive Director, Center for Health Policy and Media Engagement at The George Washington University.