Can AI Be Licensed Like Clinicians?
Generative AI is already making significant strides in the healthcare sector, assisting clinicians in interpreting lab results, dosing antibiotics, and responding to patient queries. However, current approval processes for health AI rely on the Food and Drug Administration (FDA)‘s software-as-a-medical-device (SaMD) framework. This framework is most effective for tools designed for narrow tasks, such as classifying high-risk lung lesions. In contrast, modern generative AI products possess a broader skill set, often built on foundations created by one company, fine-tuned by another, and extended by third-party plug-ins, leading to a lack of clear regulatory accountability.
Proposed Licensing Approach
In a recent perspective piece published in JAMA, Senior Fellow Eric Bressman and co-authors propose a transformative approach: licensing AI systems similarly to how we license human physicians, nurse practitioners, and physician assistants. These professionals practice in a supervised and collaborative manner, which could be mirrored in the regulation of AI.
The authors argue that concerns surrounding generative AI—such as hallucinations and performance drift—echo historical worries from the late 19th century regarding quack remedies and inconsistent clinician training. The licensure model, which combines practice standards with ongoing surveillance and education, can be adapted to regulate AI effectively.
Framework for AI Licensing
Bressman and his colleagues propose the establishment of a new federal digital licensing board to oversee AI regulation. They also recognize that existing federal and state bodies could play pivotal roles:
- The FDA could retain its role in premarket assessments, thus preventing developers from having to navigate 50 state licensing authorities.
- Designated health systems with AI expertise could act as implementation centers.
- Continuing oversight and disciplinary actions could be managed by state medical boards, with a federal coordinating body to harmonize standards.
As the authors state, “A licensure framework may help ensure that innovation scales with accountability and not ahead of it.”
Proposed Parallels Between Clinician and AI Licensing
The authors present a table outlining parallels between human clinician licensing and a potential AI licensing structure:
- Licensure Concept
- Human Clinician: Accredited degree, passing national board examinations, supervised clinical training period.
- Generative AI: Technical validation for predefined competencies (AI national board examinations), supervised pilot in nationally accredited “implementation centers” (AI “residency”).
- Scope of Practice
- Human Clinician: Delineation of approved medical services, populations, and degree of autonomy; collaboration or supervision agreements for PAs/NPs.
- Generative AI: Delineation of approved functions (e.g., image interpretation), populations, and degree of autonomy; guidance on supervising clinician oversight for each function.
- Institutional Credentialing
- Human Clinician: Health systems credential to perform specific procedures and can suspend privileges for safety concerns.
- Generative AI: Health systems’ AI governance committees vet site-specific implementation, monitor local quality metrics, and can revoke privileges if thresholds are not met.
- Continuing Oversight
- Human Clinician: Continuing education requirements and periodic knowledge assessments for board certification maintenance.
- Generative AI: Annual rerun of updated benchmarks for each competence and reporting of clinical performance measures for review by board.
- Discipline/Liability
- Human Clinician: State medical boards investigate complaints, can fine, suspend, or revoke licenses, and mandate retraining.
- Generative AI: Digital boards receive and process complaints, can place AI systems on probation, and maintain a public database of disciplined models.
In conclusion, as AI continues to evolve within healthcare, establishing a licensing framework akin to that of human clinicians could provide a structured approach to ensure both innovation and accountability.
