ChatGPT: Policy Brief
AI-Assisted Prescribing and the Transition Beyond Traditional Medical Specialisation
Executive Summary
Artificial intelligence (AI) has already surpassed human specialists in some of the most complex areas of medicine, particularly medical imaging (Radiology). Despite this, prescribing medications remains restricted to traditional specialist-led models that rely on outdated assumptions about expertise and safety. This policy brief argues that AI-assisted prescribing, supervised by licensed physicians (Medical Facult graduate MD, General Practitioner) rather than specialists, is both feasible and necessary. Recent legislative developments in the United States demonstrate that regulatory systems are beginning to recognize this shift (Contemporary Health, 2025). Also USML comparable but open book, open AI exams are needed.
The key challenge is no longer technological readiness, but institutional inertia. This brief proposes a phased regulatory framework for AI-assisted prescribing, reform of medical licensing exams, and a redefinition of clinical roles aligned with real-world practice.
Background and Context
Medical specialisation historically relied on selective memorisation and information filtering. Large volumes of knowledge were learned and subsequently discarded, with emphasis placed on exam relevance and dominant guidelines. While this model once served constraints of human cognition, it is no longer optimal in an era where AI systems can retain, cross-reference, and audit vast medical knowledge continuously.
AI has already demonstrated superior performance in radiology, including early detection of malignancies, haemorrhage flagging, and automated prioritisation of urgent cases. Prescribing medications, by comparison, is a more structured, rule-based, and auditable task for AI. Therefore, restricting AI from prescribing while accepting its dominance in imaging lacks scientific coherence.
Current Legislative Developments
In the United States, a bill introduced to Congress proposes amending the Federal Food, Drug, and Cosmetic Act to allow AI systems to prescribe medications, provided they receive state authorisation and FDA approval (Contemporary Health, 2025). While some medical specialists have expressed concern regarding safety and regulation, even critics acknowledge that AI prescribing is likely inevitable if properly governed.
This legislative effort reflects a broader global trend: technology is advancing faster than regulatory frameworks, creating a gap between clinical reality and policy.
Problem Statement
Current prescribing models face several systemic limitations:
-
Over-reliance on specialist authority rather than auditable performance
-
Limited access to care due to specialist bottlenecks
-
Lack of transparency in clinical disagreement and decision-making
-
Licensing exams that assess memorisation rather than real-world wide competence
These limitations contribute to inefficiency, inequity, and avoidable clinical risk.
Policy Proposal: AI-Assisted Prescribing Model
This brief recommends the adoption of a multi-layer AI-assisted prescribing framework, centred on licensed physicians rather than traditional specialists.
Core model:
-
Two independent AI systems generate prescribing recommendations
-
Approximately 5% of cases show divergence between AIs, treated as safety alerts
-
A licensed physician (general practitioner (Pratisyen Hekim, Tıp-Doktoru, Tabip), a graduate of Medical Faculty or internist) reviews discrepancies between 2 AIs.
-
A third AI system conducts automated auditing (dose, interactions, contraindications)
-
Final responsibility and sign-off remain with the physician
This model increases transparency, reduces fatigue-related errors, and makes disagreement visible rather than implicit.
This model eventually, maybe in 15 years from now could be automated led by AI.
Role Redefinition in Clinical Practice
-
Licensed physicians remain essential for patient context, ethics, and accountability
-
Traditional specialists are no longer required for routine prescribing
-
Researchers and clinician-scientists remain indispensable for knowledge generation, trials, and innovation
Clinical decision-making and scientific discovery must be clearly separated in policy design.
Reforming Medical Licensing and Assessment
To align regulation with real-world practice, medical licensing exams should evolve:
-
Retain foundational exams (e.g., USMLE)
-
Introduce a mandatory AI-assisted, open-book prescribing examination AI-USML
-
Require a medical degree for eligibility
-
Assess AI usage, error detection, justification of decisions, and response to risk alerts
If an AI-generated recommendation is unsafe and not corrected by the physician, scoring penalties should apply. Responsibility remains human.
Risk Management and Safety Considerations
-
AI prescribing systems must be FDA-approved and state-authorised
-
Medication classes may be introduced incrementally based on risk level
-
All AI outputs must be explainable, logged, and auditable
-
Continuous post-market surveillance should be mandatory
This framework prioritises patient safety while enabling innovation.
Implementation Timeline (Indicative)
-
0–2 years: Pilot hospitals, low-risk medications, regulatory sandbox
-
3–5 years: Expanded medication classes, national audit standards
-
5–10 years: Full integration into routine care
-
10–15 years: Gradual reduction of any human oversight as systems mature
Conclusion
AI-assisted prescribing is not a speculative future development. It is a logical extension of capabilities that already outperform human specialists in more complex domains. Continuing to restrict prescribing to traditional specialist models does not enhance safety; it preserves outdated structures.
Policy should focus on performance, transparency, and accountability, not professional preservation. With proper regulation, AI-assisted prescribing supervised by licensed physicians can improve access, even in 2 years, reduce errors, and modernise healthcare delivery.
Reference
Contemporary Health. (2025). US bill proposes allowing AI to prescribe medications. Contemporary Health – Digital Health.
