Minnesota Medical Association AI Whitepaper
Minnesota Medical Association (MMA) Task Force report on AI in Healthcare:
Background: AI adoption in healthcare is outpacing regulation. There's no comprehensive federal or Minnesota-specific governance structure — the FDA only regulates AI tied to medical devices, while large categories of clinically relevant AI (like many clinical decision support tools) fall outside its authority. Disjointed regulation at the federal level has shifted the onus of AI regulation to the states. As of Mid 2025, 17 states had enacted healthcare -specific AI legislation.
MMA AI Task Force Process: The task force, which was comprised of physicians from various medical specialties, worked though four core topics: Transparency, Bias, Liability and Clinical Decision Making.
Definitions Adopted:
Artificial Intelligence (AI): computers performing tasks typically associated with human reasoning.
Automated decision making: A type of AI in which data and algorithms are used to make decisions without human intervention.
Algorithmic/data bias: Prejudices in favor of or against a person, thing, or entity. Algorithm bias occurs when there is an underlying problem or flaw with the algorithm used to deliver outputs. Data bias occurs when the data used to train AI systems is biased in some way.
Black Box: The inability of a user to understand the specific steps taken by an algorithm that lead to an algorithm’s final output.
Foundation Model: Models trained on large datasets – and thus broadly applicable – and can be adjusted for specific applications.
Generative AI: AI systems that are capable of generating novel text, images, videos, or other outputs, typically based on foundation models. Foundation models are models trained on large datasets – and thus broadly applicable – and can be adjusted for specific applications.
Machine Learning: A subtype of AI in which complex algorithms are trained to make predictions about future outcomes. Machine learning can be supervised or unsupervised.
Transparency: Refers both to the ability to access information about an AI model’s training data and model details as well as the disclosure and documentation of the utilization of AI in health care decision-making
Six Core Issues and MMA Recommendations/Policies
AI is evolving rapidly and physicians must understand and adapt to the changing
landscape.
State sponsored advisory committee to continually review and improve AI regulation in healthcare.
Inclusion of comprehensive education at all levels of physician training, from medial school to continuing professional development.
Develop comprehensive framework for use of AI in healthcare.
MMA will monitor AI regulations that impact healthcare and educate its members on their impact.
Individuals have the right to transparent, honest and timely information about their healthcare
Physicians have a responsibility to educate themselves on AI enabled tools utilized in their practices
Disclose AI use that affects clinical decisions or patient care
Disclose AI-generated clinical communications that have not been reviewed by a physician
Specialty societies should educate their members on specific AI tools
AI developers should provide detailed information about an AI tool, including training data, data collection practices, and risk and discrimination mitigation strategies
Hospitals and facilities should share AI tool information with their clinicians
Patients have a right to unbiased healthcare
Support equitable AI access
AI may exacerbate existing social inequities.
Account for bias in AI model design, training, and use to help mitigate biases.
Continuously evaluate of AI models for bias.
MMA expects that immediate action will be taken to correct or mitigate encountered bias.
AI should not replace physician judgement
Healthcare professionals always make the final decision regarding the provision of healthcare.
MMA opposes insurance processes that utilize AI to deny coverage for a healthcare service without the input of a healthcare professional.
MMA opposes requirements from payors, hospitals, health systems, or governmental entities mandating the use of an AI tool for clinical decision making as a condition of licensure, participation, payment, or coverage.
Liability should fall on whoever is best positioned to prevent harm
The MMA supports policies that hold physicians liable only for their own clinical decision-making, not the AI tool itself.
MMA encourages multidisciplinary governance structures within systems that monitor and evaluate the performance and use of AI tools within their system