Assuming the next three years resemble the last three, GenAI tools will be eight to 16 times more powerful, capable and clinically useful than they are today.
getty
Imagine if the only way Americans traveled was by foot or horseback. And assume that consequently 100,000 people died each year because they couldn’t get to a hospital in time or firefighters arrived too late.
Then, suddenly, a technological breakthrough makes cars and trucks widely available within three years.
How would the country respond?
I predict that leaders would not wait until cars were 100% risk-free. They would move quickly to build roads, accelerate production and help people learn to drive. They would recognize that although cars would cause some accidents and deaths, fast transportation would save far more lives than it cost.
That is how America should think about generative AI in medicine.
The Status Quo In Medicine Is Anything But Safe
Most Americans know healthcare is too expensive and too hard to access. Far fewer understand how many people die each year from poorly controlled chronic disease, misdiagnoses and preventable medical errors. Fewer still can imagine how large language models like ChatGPT, Claude and Gemini could help prevent those deaths.
Today, chronic diseases like hypertension and diabetes remain poorly controlled for millions of Americans, contributing to as many as 50% of all heart attacks, strokes and kidney failures according to data from the Centers for Disease Control and Prevention. Diagnostic errors are estimated to kill or permanently disable nearly 800,000 Americans each year based on research compiled by =the Johns Hopkins Armstrong Institute Center for Diagnostic Excellence. A survey from Notable found more than 60% of patients skipped a doctor visit in the past year because scheduling was too much of a hassle, and nearly half of U.S. adults worry they cannot afford needed care in 2026, according to a recent West Health-Gallup survey.
Over the next three years, generative AI could help prevent hundreds of thousands of these deaths while improving access and outcomes for millions of Americans.
The Lifesaving Trajectory Of Large Language Models
When the first generative AI tools reached the public in November 2022, the technology was deemed remarkable but unreliable, and far from safe in clinical settings. In late 2022, Google’s first medical AI model, Med-PaLM, achieved a mere passing score of 67% on the U.S. medical licensing exam. Researchers, clinicians and regulators quickly urged caution. A mere four months later, Med-PaLM 2 scored at an expert doctor level of 87%, according to data released by Google.
Assuming the next three years resemble the last three, GenAI tools will be eight to 16 times more powerful, capable and clinically useful than they are today.
And today’s models are already impressive. A new Harvard-led study, conducted with Stanford collaborators and published in Science, tested OpenAI’s o1 preview model on difficult clinical tasks, including 76 real emergency-room cases at Beth Israel Deaconess Medical Center in Boston. At three stages of care — initial triage, first physician contact and admission to the hospital — the model matched or exceeded the performance of experienced physicians on text-based diagnostic and clinical management tasks.
And yet, physicians, elected officials and policy experts remain deeply skeptical about allowing these tools to provide direct medical guidance to patients without physician oversight.
This is a typical response to new technologies that improve exponentially. People tend to judge them by what they can do today rather than what they will soon make possible. It’s called “snapshot bias,” and it continues to shape the debate over generative AI in medicine.
To save lives, our nation must approach today’s technology not as the endpoint, but as its beginning.
The next life-saving opportunities are already within reach. Existing models are being trained on richer clinical information: conversations between doctors and patients and millions of medical records. GenAI is also capable of mining and analyzing the “estimated 900,000 data points of patient data on a critical care bed are thrown away per patient per hour,” according to Beth Andrews at Dell Technologies, who added that “97% of the data today in healthcare is not used” because it has been too difficult for humans to wrangle.
And yet snapshot bias and fears about GenAI’s impact on the medical profession stand in the way. If American medicine were doing a great job today, we could afford to slow-play AI implementation. But given the hundreds of thousands of lives lost annually from medical failures, the safer choice is to move our foot from the brake to the accelerator.
How To Accelerate GenAI In Medicine
To move fast enough, the nation should focus first where medicine is failing most and GenAI can have the most positive impact. Three opportunities stand apart:
1. Continuous Chronic Disease Control
The CDC estimates that conditions like hypertension, diabetes, heart failure and kidney disease affect 75% of American adults and drive many of the nation’s deadliest complications, including heart attacks, strokes and kidney failure.
Yet despite effective medications and evidence-based clinical protocols, these chronic diseases remain inadequately controlled.
Two major reasons are medical structure and clinician time. Chronic diseases are managed episodically, with office visits every three or four months. Meanwhile, physicians have too little time during these visits (a JAMA Health Forum study found the median primary care physician spent an average of 18.9 minutes with each patient) to evaluate trends, adjust treatment and ensure patients understand what to do next.
A generative AI application connected to a blood pressure cuff, glucose monitor or wearable device would give patients continuous insight into whether their condition is improving, worsening or drifting into dangerous territory. Patients whose readings improved could continue their current plan, while those whose conditions worsened would have the information sent to their doctors for timely medication adjustments.
To effectively control chronic conditions, all Americans will need access to GenAI, along with training on how to prompt large language models and ask follow-up questions. Wearable and bedside monitors will need to connect to GenAI applications for continual analysis. Medical societies and clinicians will need to design the analytic tools, decide when the clinician should be notified and work with researchers to confirm the reliability of the clinical pathways.
2. Medical Guidance Beyond Office Hours
When people develop symptoms at night, on weekends or between visits, they have long faced three bad options: wait until the doctor’s office opens, go to an overcrowded emergency department or search the internet and hope the information is trustworthy.
GenAI offers a better option: personalized guidance that helps patients understand the most likely diagnosis, the appropriate treatment and whether they should seek immediate medical care.
Medical societies and clinicians can help by creating educational tools that help patients safely and effectively use generative AI technology. And recognizing that these large language models will identify serious medical concerns, they can also implement systems in which telemedicine clinicians are immediately available to provide assistance.
3. Clinical Support That Gives Doctors More Time
Most misdiagnoses and preventable medical errors happen not because clinicians lack knowledge, but because they lack the time and support to apply it consistently.
During brief and infrequent office visits, they are expected to evaluate new problems, manage chronic conditions, address preventive care and comfort anxious patients. It’s not possible.
Technology companies are already building AI tools to reduce administrative burden. That is useful, but not revolutionary. The bigger opportunity for GenAI is clinical: providing care for patients with straightforward acute problems, monitoring chronic diseases, informing clinicians about potential medial errors as they arise and serving as a real-time colleague and second opinion.
Success will require medical societies, clinicians and technology companies to collaborate. The goal won’t be to replace clinicians with GenAI. It will be to complete clinical tasks that do not require a physician’s full expertise. Continuous chronic disease monitoring, 24/7 access to clinical expertise and more time for doctors to handle complex judgment would be powerful first steps.
The biggest hurdle will likely be financial, not technical. In today’s fee-for-service system, many tasks GenAI could perform are financially rewarding when done by a clinician. To align incentives, insurers, payers and self-funded employers will need to modify reimbursements in ways that encourage clinicians to prevent disease, avoid complications and reduce unnecessary hospital admissions.
Instead of viewing GenAI mainly as a tool to decrease head count, payers should recognize the hundreds of billions of dollars that could be saved through better health and fewer life-threatening medical problems.
The Safest Path Is Forward (And Faster)
Rather than seeing American healthcare as a system short on people and time, we should recognize the huge opportunities in front of us. Until now, when discussing healthcare’s core goals — quality, access and affordability — the best we could do was pick any two. GenAI can change that. Superior quality and convenient access can drive affordability for patients and the nation.
The first step is to acknowledge the preventable deaths caused each year by today’s medical system. Then we can embrace the dozens of ways that GenAI can support clinicians, empower patients and save lives.