June 19, 2026
Dear Interested Readers,
AI May Not “Save” Healthcare, but It Is Changing It
If you follow the list of headings that appear at the beginning of each of these letters, you know that one of the most frequent topics is “Future of Healthcare.” It is also true that Triple Aim, Social Determinants of Health, Politics and Healthcare, Universal Access, Burnout, Moral Injury, and Workforce Shortages in Healthcare are also topics that I frequently bring into these discussions. It occurs to me that, even though AI in healthcare can be considered a standalone topic, it is also likely to be woven through almost any current healthcare topic I might feel the need to discuss.
I have reported that I have enjoyed using ChatGPT and Google’s Gemini to organize some of the complex considerations arising from any attempt at an in-depth examination of the concerns we face. I have found AI especially helpful in reviewing the history of some of our current problems, such as how we evolved into our current physician shortages, which so significantly limit access to care.
My years as a clinician allowed me to personally view almost the entire evolution of computers in health care from “read-only” replacements of the paper medical record in the late sixties to the evolution of enhanced interactive tools that feature clinician-patient communication through “patient portals,” but have also been a major component of clinician burnout and patient complaints that the computer has come between them and the attention they need from their doctor. I don’t think I am the only patient ever to be irked by trying to have a conversation with a doctor who seems more focused on typing into a computer than listening to my concerns.
Dr. Robert Wachter, Professor and Chair of the Department of Medicine at the University of California, San Francisco, wrote quite effectively about the digital evolution in healthcare in his 2015 New York Times bestseller, The Digital Doctor: Hope, Hype, and Harm at the Dawn of Medicine’s Computer Age. Despite the fact that I had been fortunate enough to spend my career in a practice that had been using a “computerized” medical record for more than forty years, in 2009, when Barack Obama signed the HITECH Act (Health Information Technology for Economic and Clinical Health Act) into law as part of the broader American Recovery and Reinvestment Act (ARRA) over 80% of American doctors were still using paper records. The bill invested at least $30 billion in digitizing American healthcare. That reality explains the subtitle of Dr. Wachter’s 2015 book, Hope, Hype, and Harm at the Dawn of Medicine’s Computer Age. I heard Dr. Wachter speak about his book at the annual meeting of the Institute for Healthcare Improvement in Orlando, Florida, and drew on that exposure to discuss it in my December 31, 2015, letter. In that letter I wrote:
Dr. Wachter began his speech at the IHI by pointing out that because of the 30 billion dollars spent by the government, the use of computers in medicine had skyrocketed from 10% of offices and hospitals in 2010 to 75% in 2014. Those of us who have been using computers in healthcare since the late 60s have been joined by a stampede of new users. By nature, I am a little shy of being an “early adopter,” but I do get into new technology sooner than many people. I first started using email in the mid-nineties and immediately recognized that putting my email address on my business card in the office would create a closer connection between me and my patients. Harvard Community Health Plan opened its doors in October 1969 with its medical records on a computer. We had been on Epic since 1995 and have had a patient portal for the last decade.
These days, with a majority of physicians being employed by for-profit and non-profit healthcare systems, it is rare to encounter a clinician who isn’t “digitized.” While many practices were struggling to get “digitized,” at Atrius Health we were developing a “data warehouse” that allowed us to be leaders in the “quality movement,” but one of my most disappointing realizations as the chair of our board of directors in the early 2000’s was to discover that the tool we were developing would be of more use to our CFO than to our CMO. I think many clinicians would agree that giving patients digital access to their medical records and providers’ office notes was a noteworthy accomplishment with a mixed, transformative impact on practice. It was great to give patients access to their medical records, but it did not necessarily improve the conversations between concerned patients and their doctors and nurses. The combination of easy access to the Internet and the ability to peruse their medical records and test results led to a change in the dynamics of doctor-patient interaction, as many routine office visits were transformed into long sessions spent explaining how apparent abnormalities did not map onto information and misinformation gleaned from Google. Suffice it to say, we have had a remarkable tool that not everyone has yet learned how to use, and Dr. Wachter saw it coming.
Earlier this spring, I quickly burned through both the digital and audio versions of Dr. Wachter’s new book, A Giant Leap: How AI Is Transforming Healthcare and What That Means for Our Future. Dr. Wachter treats this extension of the story of healthcare’s IT evolution as a sequel to his original book about computers in healthcare, and begins his story (much of the book is about his experience testing the utility of AI in his own teaching and practice) by reviewing the digitization of healthcare, which essentially began in earnest with the opening created by the Obama administration’s investment back in 2009.
Sometimes, all you need to read in a non-fiction book to get the impact is the preface. Wachter’s prefaces are exceptional essays on his subject. He ends his lengthy preface by writing:
There will inevitably be sharp growing pains as we integrate AI into healthcare, and some patients will be harmed, maybe even killed, as we sort out its proper place. But, in the words of Joe Biden, “Don’t compare me to the Almighty; compare me to the alternative.” Our current healthcare system fails patients and clinicians all too often and is unsustainably expensive, inaccessible, unwieldy, and infuriating. That means that AI doesn’t have to be perfect to be better. Which is good, since it won’t be perfect. The key question is whether it will truly be better.
Flipping the page to Chapter I, which is entitled “An Overnight Revolution, Fifty Years in the Making: Lessons of the EHR Era,” we read:
In Ernest Hemingway’s classic 1926 novel The Sun Also Rises, a character named Mike Campbell loses all his money due to a series of reckless financial choices. “How did you go bankrupt?” his friend Bill asks. “Two ways,” Campbell replies. “Gradually and then suddenly.” When it comes to the digital transformation of healthcare, we have the “gradually” part down pat—no industry has been slower than healthcare in disrupting the status quo with digital tools. Could the introduction of generative AI be our “suddenly” moment, when a breathtaking new technology crashes into a healthcare system in desperate need of change, igniting true transformation? Answering this question requires some appreciation of the history of healthcare digitization.
What follows is what I have already implied. Healthcare was slow to adopt the benefits of computers. Banking and other financial industries had been using digital tools long before the majority of physicians and hospitals began moving from paper charts to screens. Whether due to confidentiality concerns, widespread inability to type, or simply financial constraints, most healthcare organizations were “late” adopters. I know that when my practice moved to Epic, which required physicians and nurses to type, from an electronic medical record where doctors and nurses made notations on a paper inputting sheet or dictated their notes for a medical transcriptionist to type into the electronic record, we realized that we had to offer typing lessons to our staff. Like many of my colleagues, I could not type in the mid-nineties. In high school, after a few frustrating days in typing class, I reasoned that learning to type was an activity for future secretaries, and since I planned to always have a secretary, I convinced my academic advisor that since I planned to be a doctor, a better use of my time would be to replace my typing class with an elective anatomy course, where we dissected a cat. In college, I had paid someone pennies a line to type papers for me when a typed product was required.
In time, the error of my teenage analysis became an increasing pain and a persistent problem for me. As we were transitioning to Epic, and I was typing by the “seek, and you shall find” method, my poor judgment persisted. I was too busy to take the typing class we were offering to everyone who needed help learning to type. I tried to save time and avoid typing by getting Dragon voice recognition software to work for me. The problem was that, with the inadequate computing power of most 90s laptops, Dragon did not work well, and I was never able to organize my thoughts effectively while dictating. I eventually developed my own clumsy, very slow typing methods. Years later, after becoming CEO, and after great improvements in voice recognition software, when I was no longer doing much typing into a medical record, and had a very efficient administrative assistant who could type up a storm, I did insist that we install voice recognition software as an option for entering data into our Epic medical record system. Ironically, the most avid user of our voice recognition software that I was aware of was a colleague who could type 120 words a minute but had carpal tunnel syndrome.
These days, in almost every edition of a major newspaper, in print and online articles in journals like the New Yorker and the Atlantic, on podcasts, and occasionally in medical journals, we are bombarded by opinions about the future impact of AI on our society. In the most ominous predictions, we are warned that the machines will wage war against humanity. More thoughtful articles worry about job losses, changes in jobs, or societal transformations that will rival the Industrial Revolution of the late 19th century or the Digital transformation that we have recently experienced. Wachter recognizes these concerns but still predicts that medical practice will be transformed at an accelerating pace. He admits that for him, his pocket access to AI has already transformed the way he sees patients, and teaches residents and medical students.
Wachter admits that the introduction of AI to medical practice has been a journey of uncertainty with some significant false starts, but points out that so far, most errors have been learning experiences. The biggest false start has been the effort to use AI as a diagnostic tool. Perhaps its biggest success has been as a scribe. Despite his review of the early failures of AI as a diagnostic tool, Wachter notes in his New York Times essay about AI entitled “Stop Worrying, and Let A.I. Help Save Your Life,” which is in essence a review of his book, that he now uses AI via his cell phone like a “curbside consult.” He writes:
We physicians have a long tradition of the curbside consult — when we bump into specialists or more seasoned colleagues in the hospital cafeteria and ask for their advice on a vexing clinical case. Over my 35 years of practice, I used to track down other doctors for a couple of curbsides during morning rounds each day.
These days, I’m getting far more curbsides, but they are not with colleagues. They’re with A.I. Sometimes I consult with ChatGPT; other times I turn to OpenEvidence, a specialized tool for physicians. I find A.I.’s input is virtually always useful. These tools provide immediate and comprehensive answers to complex questions far more effectively than a traditional textbook or a Google search. And they are available 24/7.
I don’t interpret that statement to mean that Wachter uses AI to tell him the diagnosis, but rather to help him consider ideas he hasn’t yet considered. The process of making a diagnosis is complex. It begins with gathering all available information from the patient’s symptoms, personal, social, and family histories, tests, imaging studies, and their physical exam. In a curbside consult, one doctor often presents this collection of information to a respected colleague who has been unexpectedly encountered in a hallway or cafeteria in a hospital or medical office building. I frequently both sought and provided curbside consults during my years in practice. Sometimes, a worried colleague would just knock on my office door late in the afternoon while I was reading EKGs and begin a conversation with, “What do you think about…” That is not much different than typing a question into an AI source.
I always thought curbside consults were more timely and efficient than sending a patient for a formal consult, which could take weeks to schedule. In a prepaid, capitated practice, the curbside consult saved office resources for things that could wait or were more complex and required deeper consideration. The clinician seeking a curbside consult is not usually asking their colleague to make a diagnosis for them, but rather to share uncertainties, consider what is known, and offer an opinion that may include new possibilities they had not previously considered. I wonder how this practice will change in the future. Will AI become a third voice in the conversation, or will it be the only voice? The most common cause of a malpractice complaint is “failure to diagnose,” which often arises from failing to consider all possibilities.
One area of practice where AI is being adopted with enthusiasm is as a scribe. We are at least a decade into widespread use of human scribes in hospital and office practice, and AI is quickly replacing that human activity. The financial advantages are obvious. The AI scribe doesn’t ask for a salary or benefits and doesn’t call in sick. A concern for both human and AI scribes is whether clinicians will carefully read what is written for them before signing it into the record. It isn’t likely that human scribes will begin suggesting diagnoses, but that may be a risk when AI is the scribe. Dr. Lisa Rosenbaum, the host of the New England Journal’s podcast “Not Otherwise Specified,” has raised concerns about the challenges facing primary care. In one podcast, she specifically worries about the downside of using AI-generated clinical notes. She suggests that the thought processes required to construct a clinical note are foundational to her conceptualization of the clinical problem and that her cognitive process is disrupted when AI constructs the clinical note. I agree with her that there is no better way to engage in a thoughtful approach to diagnosis and treatment than to write up the encounter. I see the problem she identifies as an extension of the “copy and paste” problems that have perpetuated historical inaccuracies in the electronic medical record, as clinicians seek ways to get through an overbooked schedule while patients crave more of their attention.
Dr. Wachter often offers a specific, progressive, and thoughtful response to the concern that AI has imperfections, is occasionally inaccurate, and could occasionally be dangerous. He writes:
Some people argue that A.I.’s imperfections mean that we shouldn’t use the technology in high-stakes fields like medicine or that it should be tightly regulated before we do. But the biggest mistake now would be to overly restrict A.I. tools that could improve care by setting an impossibly high bar, one far higher than the one we set for ourselves as doctors. A.I. doesn’t have to be perfect to be better. It just has to be better.
There is no question in my mind that, despite early failures and concerns about the overall accuracy of AI, there will be ongoing efforts to have AI provide definitive diagnostic opinions. AI won’t be ignored as a potential answer as long as clinicians are desperate for relief from burnout, and massive access problems persist even for patients with very robust commercial insurance policies. The temptation will be there, and hard to resist. A neighbor of mine just purchased a very expensive truck that can “self-drive.” As he was showing me all of the new technology, mostly cameras, built in to make sure that he was not asleep or surfing the Internet while his truck was changing lanes, passing slower drivers, and avoiding road hazards, I was asking myself whether there would be similar fail-safes built into the medical applications of AI. I fear that it will become an issue of individual integrity. We are left trying to decide how to safely use an impressive yet imperfect technology in a time of complex needs and challenges.
A recurrent concern that even worries such an enthusiastic early adopter as Dr. Wachter is what the outcome of AI implementation in an age of corporatization of practice will be for clinicians’ work lives. As AI does more of both the “grunt work” and diagnostic thinking in practice, will the corporate “overlords” add more visits to schedules, or will the reclaimed time be devoted to more humanly scheduled appointments that allow more time for thinking about and interacting with patients? What do you think? I think we will see some mixture of those potential outcomes.
In his New York Times essay, Wachter implies that, despite the many concerns others might have about current and future problems in the evolution of AI in healthcare and what we don’t and can’t yet know about how it could be misused, he is still in favor of taking bold steps now. Wachter sees AI as a source of healthcare transformation and summarizes what he used many pages of his book to consider:
A.I. can support this transformation, but only if we stop disproportionately focusing on rare bad outcomes, as we often do with new technologies. While research now demonstrates that driverless cars are safer than those with human drivers, a serious accident involving a robotaxi is deemed highly newsworthy and often cited as a reason to take driverless cars off the road, whereas one accident involving a human driver may hardly leave a media ripple.
Wachter’s book takes on almost every conceivable concern you might have about AI. He frequently comments that he is engaged in an advisory role for companies that produce AI apps for various practice and back-office activities. He spends time discussing the tension between the two largest medical record systems, Epic and Cerner. Cerner was purchased by Oracle in 2022 for $28.3 billion. Both companies are building AI tools into their platforms, and their users will need to decide whether to accept the AI tools offered by these two dominant players in healthcare IT or to add specialized products from other developers into their IT platforms. My response to this question is that it looks like there will be a lot of AI in our medical futures. The question is not whether AI will be used, but who will develop, supply, and improve the tools.
AI could be of great help in resolving the pain of “prior authorization.” Many clinicians are drowning in a swamp of unanswered patient requests in their “in boxes” attached to patient portals. Can “in box” management be safely delegated to an AI tool? Can AI be effectively used to read and summarize complex medical charts? Dr. Wachter sees both things happening soon, but fears that clinicians won’t do adequate reviews of what the machines offer. The question is how the partnership between AI and the individual clinician will work. It is easy to imagine that there will be great variation in how clinicians partner with AI. Core to the tension between safety and efficiency will be the need to decide, for many different tasks, how involved the clinician needs to be, or how independently the AI tool can be allowed to function. Dr. Wachter has the same answer to most of these questions. He suggests that eventually we will accept AI’s ability to do the task with a high degree of efficiency and effectiveness, not perfectly, but possibly better than humans do them now. It’s about developing trust.
As A.I. becomes more commonplace in health care, we need to develop strategies to determine how much to trust it. As we measure error rates and harms from A.I., we need frameworks to make apples-to-apples comparisons between what human doctors do on their own today and what A.I.-enabled health care does tomorrow. In these early days, we should favor a “walk before you run” strategy, starting with using A.I. to handle administrative paperwork tasks before focusing all our energy on higher-stakes tasks like diagnosis and treatment.
Dr. Wachter comes to the same conclusion in both his essay and his book. He uses AI more and more as it gets better and better. How effectively it solves the big problems that diminish the joy of practice depends on its application. It will change how we deliver and receive care, but that has been changing now for several decades. Medical students are being trained to work in a world of AI-enabled healthcare. Whether AI will make everything better remains to be seen, but that things will change is now a certainty. Wacheter remains positive. He ends his essay by once again arguing that AI doesn’t need to be perfect, just helpful and better.
But as the saying goes, “Don’t compare me to the Almighty; compare me to the alternative.” In health care, the alternative is a system that fails too many patients, costs too much and frustrates everyone it touches. A.I. won’t fix all of that, but it’s already fixing some of it — and that’s worth celebrating.
There are days when I wish I were twenty years younger and still in practice. One of the great joys of my journey through practice was to be part of the effort to improve care delivery. I am certain that it would be exciting to go to work each day, trying to use a new tool like AI in new ways to improve the care experience. Now I will just enjoy watching the evolution from the sidelines and trying to keep up with all the articles that appear in the ongoing conversation about what to embrace and what to fear.
Just to give you an opportunity to launch your own exploration, let me offer two more interesting articles for your consideration. Both are written by physicians. The first was written by Dr. Dhruv Khullar and appeared in the New Yorker last September. Dr. Khullar explores the diagnostic abilities of AI compared to our best clinicians. It is a little reminiscent of the late-90s chess battle between chess master Gary Kasparov and IBM’s Deep Blue. Deep Blue won. The article is entitled “If A.I. Can Diagnose Patients, What Are Doctors For?: Large language models are transforming medicine—but the technology comes with side effects.
A second article, also written by a physician, but describing her experience as a patient, recently appeared as an opinion piece in the New York Times. The piece, written by Dr. Helen Ouyang, was titled “Doctors, This Is Why Our Patients Are Using ChatGPT.” Dr. Ouyang got more help and emotional support for her own medical problem from ChatGPT than from her personal physician. Near the end of her piece, Dr. Ouyang writes:
It’s a grim fact of American medicine today that doctors can’t come close to a chatbot’s availability. And when the health care system can’t reliably offer time, attentiveness and compassion, patients will go searching for them somewhere else, even from a machine we assumed could never feel human. A.I. may not replace doctors, but it will change what patients expect from us. Doctors need to adapt.
Dr. Ouyang says that doctors need to adapt. I expect they will. The last 60 years of medical practice have been an ongoing exercise in implementing new technologies that have transformed care at an ever-accelerating pace. In retrospect, it has been an exciting time. It is my hope that what lies ahead as AI is incorporated into the delivery of care and the business of medicine will generate even greater benefits for both patients and practitioners. I have more concern that we won’t maximize the potential benefits of AI than that AI will, in some way, be a source of error and injury. It should be considered a tool, and our task is to continue exploring how to use it effectively.
Juneteenth, the Summer Solstice, a Bear, and Family
On this day in 1865, the enslaved people in Texas learned that on January 1, 1863, President Lincoln had signed the Emancipation Proclamation, freeing all enslaved people living in territory controlled by the Union Army. The Civil War had ended in April, and Lincoln was already in the grave before the Union Army, and the word of the end of slavery got to Galveston, Texas, 161 years ago. For many years, the former slaves in Texas remembered and celebrated the day as Juneteenth Day. I first became aware of the holiday commemorating the good news when I took Texas history in the seventh grade in 1957. I am very pleased that even though it took a long time for the celebration to spread, first through the African American community and then to other communities, it finally became a federal holiday in 2021. The celebration should be coupled with awareness that the destination of the journey toward equality is still in the distance, somewhere down the road. It is good that the celebration has spread from Texas to become a federal holiday, allowing us to reflect on what a long, long road and difficult trip it has been and continues to be.
The Summer Solstice is Sunday, but we have had summer weather for at least two weeks. I will make no more references to winter weather that won’t go away. Summer brings joys and challenges, and the challenges seem to be increasing. For example, it seems we have more ticks now than we did ten years ago. Maybe it is the same number of ticks, but the ticks have gotten a better press agent.
Another growing warm-weather problem, beyond the increased heat and violent storms, is that it feels like there are more bears. One of the things I like about where we live is that we are very close to forests. Indeed, about half of our land is heavily wooded. There were even more trees before we enlarged our front yard. Our woods connect to our neighbor’s, and his connect to other wooded areas. There is a very large wooded park in the middle of our town. Much of the shoreline around our lake is wooded conservation land where bears and other wildlife are present.
Over the thirty-plus years that I have enjoyed having a home in New Hampshire, either for vacations and long weekends or as my full-time residence, one of my greatest joys has been feeding birds. I have enjoyed having bird feeders in front of the kitchen window, and several others off the deck near the lake. One thing those of us who feed birds know is that bears like what we feed the birds. Over the years, once or twice a summer, I will get up in the morning and discover that a bear has destroyed one of the feeders. Last summer, I countered these nocturnal visits by bringing in my feeders at night, since experience suggested most visits occur then. I also developed a theory that the bears like the seeds but don’t like hot spicy suet. It turns out that all my theories are false, as you can see from the picture that is today’s header. The bear won. He took all of the hot spicy suet by the lake in broad daylight. He then came back the same night and cleaned out the feeder in front of the kitchen window. It has been sad this week for the birds and me. After I took down the suet and seed, I did not remove the poles from which I hung the feeders. I have seen some of my feathered friends sitting on the poles, like people at a restaurant table, waiting for the waitress’s attention and wondering what will be on the menu.
Part of what makes summer so great is having visitors. I won’t have birds, but all of our children and their families, along with a couple of friends and romantic partners, are coming at the same time, over the last few days of June and early July. We are expecting at least 14 to join us! It will be a multi-day party. I hope that you are looking forward to some special days around the Fourth.
Be well,
Gene
