Applying Clinical Licensing Principles to Artificial Intelligence

In this issue of JAMA Internal Medicine, Bressman and colleagues propose an intriguing idea: what would happen if medical tools incorporating artificial intelligence (AI) were licensed like advanced practice clinicians, rather than regulated solely by the U.S. Food and Drug Administration (FDA)

This strategy is intended to provide an alternative or complementary pathway to the FDA approval process for regulating AI-enabled medical software. The authors argue that such an approach could offer the flexibility needed to keep pace with the rapid evolution of AI, the broad range of applications each model may serve, and the requirement that these tools demonstrate clinical value.

While many specialized, single-purpose AI applications can be adequately regulated within existing frameworks, generative AI may be deployed across multiple contexts, and its models can continue to evolve over time. Because these models are probabilistic rather than deterministic, they may commit errors similar to human errors—such as those arising from incomplete knowledge or inaccurate judgment. Bressman and colleagues suggest that an appropriately flexible framework for certification already exists in the form of the licensing and oversight system used for advanced practice clinicians.

Image for illustration

With this approach, the level of supervision would depend on the specific activity, with some tasks requiring closer oversight than others.

This proposal, however, leaves many important details unresolved. Any licensing system for AI would need the ability to evaluate and address the specific risks posed by each model before deployment; thus, a central regulatory authority would likely still be necessary. In addition, determining who is responsible for and oversees the decisions and treatment plans generated by AI—and who holds legal accountability when errors or adverse events occur—remains a challenging question. These considerations mirror longstanding issues in clinical licensing, but although medical boards are well-positioned to grant licenses, it is uncertain to what extent a similar system could be developed with the expertise required for AI in medicine.

A licensing framework for AI in clinical practice, analogous to that used for physicians, could offer a more flexible solution than current regulatory processes and allow for ongoing evaluation and updates. The broad scope of these emerging technologies may compel those responsible for ensuring their safety and effectiveness to adopt an equally expansive and adaptable mindset.

See the full article here