In this comprehensive report, the AMA and Manatt Health explore the transformational potential of augmented intelligence (AI) in the practice of medicine and outline practical considerations for physicians who are using or considering using AI-based tools for clinical or administrative purposes.

AI in health care is not new. The use of AI in medicine dates to the mid-20th century and has been increasingly used in recent years in radiology, cardiology and neurology. Recent technological developments—such as advances in deep learning and the development of foundation models—have dramatically expanded the potential use cases of AI in the delivery and administration of health care. AI models have been used to develop cancer prognoses, respond to patient messages, predict adverse clinical events and recommend optimal staffing volumes. Just last year (2023), an AI large language model algorithm successfully passed the multiple-choice section of the United States Medical Licensing Examination—just one example of AI’s vast and rapidly expanding capabilities.

This report aims to explore the promises and risks of AI in health care and provide an overview of the key concepts, use cases, and challenges physicians should be aware of as AI tools and their use grow across the industry. This report also highlights the AMA’s commitment to advancing the safe, effective, transparent and equitable use of AI tools and services through the development of AI principles, advocacy for appropriate federal and state regulations and oversight, and physician education. The findings and opportunities outlined in this report were gathered through a survey of over 1,000 physicians, a series of interviews with artificial intelligence experts, a roundtable discussion with specialty society representatives (see Acknowledgements in report), and a review of existing literature about health AI.

Note: The American Medical Association (AMA) uses the term “augmented intelligence” rather than “artificial intelligence” to reflect its perspective that artificial intelligence tools and services support rather than explicitly replace human decision-making.