About half of the primary care visits I see these days involve patients that have consulted ChatGPT and have “an answer” to their medical symptoms that I need to debate, but that is just the beginning of what is coming to engulf all of health care in the near future. Working part-time for an Amazon One Medical clinic in Silicon Valley, I am seeing the rollout of artificial intelligence (AI) assistance take form. In January, their Amazon Health AI assistant was rolled out giving patients advice on common medical conditions, helping refill prescriptions, schedule appointments, and determine their need for care. From the provider side, most use the AI scribe which has helped significantly with reducing the burden of documentation of visits although it seems a bit generic and can be inaccurate. AI summaries of tests and radiology have become part of the medical record. I am certain they are working on a future true AI medical provider that will screen primary care appointment requests and provide diagnoses, testing, advice, and follow up. “Your AI Avatar doctor will see you now!” rings in my mind.
ChatGPT Health was also rolled out in January this year which attempts to connect your medical records into its advice service. There is currently a long waitlist to sign up. Just now getting into the game is CVS Health using agentic (planning and coordinating) AI systems in partnership with Google Cloud, although this is not yet planned to be direct patient care. All the other big players have their own AIs in the works, such as Microsoft Health, Anthropic, Apple, and NVIDIA.
The push for an artificial intelligence physician
With the growing shortage of primary care providers and no end in sight, it is inevitable that an AI physician version will come to pass. The advantages of being available 24/7, never being tired, and being consistent will likely tip the scale to significant usage. However, a recent study in Nature Medicine analyzing ChatGPT Health revealed that the results were quite concerning. While it did okay for triaging routine visits, it under-triaged patients that required emergent evaluation to an emergency room (ER) by half. This was due to its inability to accurately predict patients having a worsening condition such as asthma. Real-life evaluations still currently count for those with risky uncertain conditions. However, one big unanswered question on AI medical consultations is who is going to pay for the service and at what it will cost.
Transforming hospital systems and decision support
Looking at the entire medical care landscape, AI has already entrenched itself in many areas and is poised to transform how medical care is delivered in a way never seen before. Hospitals are increasingly using AI in almost every corner. It is already commonplace in the radiology department where it analyzes X-rays, magnetic resonance imaging (MRI) scans, and other diagnostic scans to assist the radiologist and has been proven to improve accuracy. Nearly 65 percent of hospitals use AI to assist with administrative tasks such as billing and scheduling, but newer patient monitoring systems are rapidly developing to help nurses monitor patients with electronic sensors, predict fall risk, reduce adverse reactions, and improve patient outcomes all with less administrative hassles that currently plague nurses.
For doctors in the hospital taking care of complex patients with a large data set of testing results, AI can help find the correct diagnosis and help plan treatment options. This was recently demonstrated during an interview with Robert Wachter, MD, head of the hospitalist program of the University of California, San Francisco (UCSF), and Eric Topol, MD, in a Substack interview. With this use, AI is acting as a decision support layer for clinicians.
Chronic disease management and medical breakthroughs
In chronic disease management, AI support currently has its best impact on diabetes in particular with continuous glucose measurement devices worn on the arm with minimal irritation. While the analysis and prediction components are helpful, they cannot be used as standalone without a provider’s input as of yet. In heart disease and hypertension, AI interpretation is dependent on the accuracy and consistency of the data from the patient’s wearable devices. AI analysis can also incorporate other factors such as lab and other tests to improve prediction of the course of disease and suggest management options.
Behind the scenes, AI has already made a big impact with medical science discoveries and innovations. The Nobel Prize in Chemistry was won in 2024 by two researchers of Google DeepMind for developing AlphaFold, an AI system that predicts protein 3D structures accurately. This has advanced understanding of cell microbiology astronomically and is leading to developing specifically designed drugs for use (none yet have been approved) in a wide variety of illnesses.
In the world of genomics, AI is able to analyze large data sets and begin to unravel the significance of certain variations on diseases. In therapeutics, it is being used to develop specific messenger RNA (mRNA) personalized cancer vaccines to improve treatment protocols such as with melanoma. In addition, analyzing unique genetic markers can predict which medications will work better with fewer side effects. Also, patients with rare genetic conditions can get an accurate diagnosis faster with AI analysis, and it is in development to assist genetic treatments, for example with clustered regularly interspaced short palindromic repeats (CRISPR) technology.
The reality of consumer wearables and regulation
With consumer wearables such as smart watches, rings, and wrist devices, AI has already been incorporated for some time to interpret results of data such as heart rate, oxygen saturation, basic electrocardiogram (EKG) readings, sleep, step analysis, and “recovery” analysis. Unfortunately, the interpretations from these devices have significant shortcomings in many areas. They are generally good at baseline heart rates when at rest, but the accuracy goes down with exercise due to movement and interference. Detection of a common arrhythmia, atrial fibrillation, is actually reasonably sensitive and can be used as a screening device on the Apple Watch, for example. Sleep analysis performs the worst of all the attempted measured parameters with wide and inaccurate variations for sleep efficiency, stages of sleep, and actual sleep time. Step count is reasonably accurate with walking but has been shown to generally underestimate actual steps. How much better these devices will become is unknown.
One could easily see how health care was ripe for a full-on assault by AI as it has traditionally been lagging behind for decades on adoption and development of highly effective technology. However, with medical care eating up nearly 20 percent of the gross domestic product (GDP) in the U.S., developing technologies have been able to focus on making inroads into this vast and lucrative market with AI seemingly leading the charge. While there is great promise to improve medical care delivery and improve efficiency and possibly cut costs, there is likely a downside to consider, especially with the lack of regulation and a government stance to stay away from regulating AI.
The current administration’s view of “hands-off” any AI regulation along the offices of the Food and Drug Administration (FDA) and the Centers for Medicare and Medicaid Services (CMS) adopting AI protocols quickly, we may see the effects good and bad in the near future. The FDA’s controversial commissioner, Marty Makary, has stated they will incorporate AI tools into every aspect of the agency from the application process to final approval of drugs and devices. CMS has an AI Playbook it is using to incorporate various tools for its internal workflow, authorizations, reducing fraud, advancing innovation, and, interestingly, using AI as a way to safeguard itself from over and wrongful use. The future is now and there is no turning back. Will medicine be able to accurately handle the bumps and missteps AI throws at it? Stay tuned!
George F. Smith is an internal medicine physician.
