The future is here

The AI Revolution in Medicine: GPT-4 and Beyond is a must-read for physicians both in training and practice. Written for the general populace by experts at the intersection of medicine and technology, the book provides transformative insight into the applications of artificial intelligence (AI), ranging from patient care and workflow to research and regulation.

Our biggest takeaway was that the capability and impact of AI are advancing much more rapidly than anticipated. This book efficiently brings readers up to speed. It urges readers to recognize the advanced state of machine learning algorithms, capable of instantly analyzing vast amounts of data and synthesizing it in an easily understandable way. These algorithms can swiftly parse through thousands of scientific publications, which would take a human months if not years. AI can take this a step further by using the information to make accurate predictions and suggest future steps. It can even infer the emotional states of humans and provide empathetic counseling. This starts to blur the role of the physician in healthcare.

This technology comes at a time when we desperately need it. The medical literature base is expanding by a new scientific paper every 30 s, and the rate of publication is doubling every 5 years. ¹ This is impossible for humans to keep up with. Medical-purpose AI such as OpenEvidence ² are already being developed, with specialized training on the scientific database for the purpose of translating information to researchers and healthcare workers. As the rate of progress accelerates beyond the abilities of the native human mind, integration of AI technologies may be a necessary next step to keep up with the pace.

This speed of development will leave regulatory frameworks trailing behind. The clinical application of AI is currently in a relatively lawless era, leaving us to grapple with pressing moral questions. Which applications of AI in medicine are ethically acceptable? How will transformative and evolving AI applications be controlled at the individual, institutional, and government levels? What are the potential mechanisms of corrective actions and legal repercussions? As governing policies gradually come into existence, physicians both in training and practice need to pay special attention to preserving our autonomy and “be at the table,” rather than side-lined by “special interest groups.” AI has formidable potential as a tool to help physicians deliver efficient and data-driven care, but there is potential that mismanagement might render us vulnerable to new levels of monitoring, or as a target to blame when the algorithm falters. Ethical considerations should drive the forefront of “officially-approved” AI models in healthcare and should be filtered through the minds of clinical experts. However, the enormous investment required to finance intelligent neural networks may prohibit ownership to all healthcare providers except large institutions. Support for regulatory policies that sustain open access and fair use to the general public will be key in the delicate balance between harnessing the benefits of medical AI and safeguarding the mission to provide equitable healthcare delivery.

We know that levels of physician burnout remain exceptionally high.^3,4 AI has the power to make or break us. As the authors discuss, there is great potential in AI's capacity to streamline administrative functions, documentation processes, and information acquisition. Yet, we ought to recognize that burnout and moral injury stem not just from an excessive workload but more so a diminishing sense of professional autonomy. The relentless advance of the information age and the corporatization of medicine have steadily eroded individual physician decision-making capabilities. We are well acquainted with the predicament presented by the patient who, after engaging in well-intended but inadequate internet research, arrives at the medical office armed with inappropriate treatment demands. Similarly, the challenge persists with insurance companies that repeatedly deny coverage for essential treatments, despite compelling evidence substantiating their necessity. If AI implementation goes awry, it could further supplant our expertise, reducing physicians to mere technicians dutifully executing prescribed tasks.

As a fellow pediatric subspecialist, a deep connection has been felt with the input from Dr Isaac Kohane in this book. Pediatrics is recognized for its distinctive challenges and complex diagnostic processes, areas where AI may have a large positive impact. Dr Kohane mentions the unfortunate loss of a young life that could have been prevented had the necessary innovations in critical respiratory care come into being just a year earlier. In physiatry, we value a trans-disciplinary approach in healthcare delivery from the perspective of how one's health is influenced by the existing conditions and contextual factors both personal and environmental. This is assessed mainly at three levels: the biomechanics of the individual, the interactions with the immediate physical space, and the ability to integrate into society. ⁵ In pediatric physiatry, developmental considerations as well as psychosocial caregiver dynamics are important factors in care coordination. The multidisciplinary team of healthcare, habilitation/rehabilitation, and educational professionals play unique and necessary roles, but coordination at this scale becomes difficult. In theory, properly implemented AI could facilitate communication, allowing everyone to stay in the loop. As patient-centered providers, we cannot delay this in good conscience. The answer lies in facilitating the implementation as stewards, to ensure that the priorities of patient care and quality of life are not lost.

This technology is not developing in a vacuum. It is developing in the offices of your companies, in the hospitals in which you receive care, and in the comfort of your own home. The big question the medical community should be asking is, “Will this technology free us to focus on our core duty to the people, or will it further empower administrators to dictate our actions?” Large hospital systems throughout the United States have already begun experimenting with AI incorporation into the documentation process. The scientific community is developing large language models that are updated in real-time on the entire body of scientific literature. It is already here, and it is just the beginning. Last year, a survey revealed that the average American reads only one book in a year. ⁶ If that is you, then this is that book.