Skip to Main Content

The FDA’s announcement last week that it approved artificial intelligence software that can identify diabetic retinopathy, a common eye disease, without the need for an eye specialist likely shook some doctors already concerned about this new technology. I don’t think they have anything to worry about.

Artificial intelligence, sometimes called AI, is not a dramatic and revolutionary development in the history of medicine. It’s but the latest in a long line of breakthroughs that have made it possible for caregivers to better diagnose and treat illness. We should be wary of the hype surrounding this advance, which is leading to broad misconceptions that AI will replace doctors. What it will actually do is put a premium on physicians’ knowledge and decision-making skills.


The history of medicine during the last two centuries should be seen as the development of ever more precise tools that help doctors narrow the range of possible causes for their patients’ symptoms. One of the great breakthroughs occurred in 1816 when René Laennec used a tightly rolled a sheet of paper to listen to a woman’s heart. This stethoscope helped launch a new era in medicine that placed less emphasis on patients’ descriptions of their symptoms and more on the search for clues within their bodies.

Since then, the development of a dazzling array of increasingly precise diagnostic tools — from blood tests and X-rays to EKGs, CT scans, and gene sequencing — has given doctors a much more accurate sense of the biological and chemical roots of disease.

The development of medical charts in the early 20th century gathered a patient’s information into a single file. These charts allowed doctors to better track the effectiveness of their treatments. They also helped transform medicine from an art based on personal knowledge and intuition into a science based on objective evidence. The development of electronic health records during the last few decades has vastly increased our ability to gather and analyze patient data.


In fact, during the last few decades technology has flipped the information problem upside down. Instead of having too little information, caregivers today have far more than they can process on their own.

This is the great promise of artificial intelligence: to train machines to comb through petabytes of data on their own to find patterns that would elude human caregivers. It is already paying dividends. Stanford researchers used a deep-learning algorithm to identify cancerous spots in medical images. An Israeli-based medical analytics company recently announced it had developed an algorithm capable of detecting intracranial hemorrhages, which are often missed and contribute to nearly 1 million deaths worldwide each year. And researchers at my school, the University of Michigan, have produced an algorithm that analyzes more than 4,000 distinct variables to predict who might be susceptible to contracting a dangerous intestinal infection called Clostridium difficile, or C. diff.

These and other promising breakthroughs have led some to suggest that machines may supplant doctors, just as autonomous vehicles might someday replace drivers. Such thinking puts the cart before the horse. Like the medical technologies that have come before it, artificial intelligence is another tool that will make the knowledge, skill, and judgment of physicians even more central to quality care.

The human body is such a complex and dynamic entity — each one unique in its own way — that medicine will never become a mathematical problem in which data can be crunched to produce the single right answer. It will always involve a process of elimination that allows the caregiver to focus on the most likely cause of illness and determine the most effective treatments from a range of options.

Artificial intelligence will undoubtedly inform and improve this decision making process — guided by physicians. The great challenge going forward is in recognizing and nurturing this irreplaceable human element, to train doctors to work with machines without becoming too reliant on them, and to never forget the centrality of the doctor-patient relationship and the importance of the human touch. Just as the stethoscope was only as useful as Laennec’s ability to assess what he was hearing in his patient’s chest, the success of artificial intelligence will depend largely on the ability of physicians to interpret and apply its findings.

Dr. Abraham Verghese, a graceful writer and professor at the Stanford University School of Medicine, put it best when he observed, “The way here is not to think technology versus human, but to ask how they come together where the sum can be greater than the parts for an equitable, inclusive, human and humane care and practice in medicine.”

Marschall S. Runge, M.D., is the executive vice president for medical affairs at the University of Michigan and dean of the University of Michigan Medical School. He receives funds from Eli Lilly for his work as a member of its board of directors.

  • It’s naïve to think that AI won’t replace humans. In many cases, that’s *exactly* what is designed to do. In the hands of profit-seeking corporations determined to cut the expense of human labor, replacing humans is exactly what the doctor ordered. And it will be businesses–not ethicists, scientists, or doctors–that determine how AI will be implemented in the marketplace. Self-driving cars are not designed to put a premium on drivers’ skills; they are designed to utterly wipe them off the planet. Discussing old tools or upgrades of old tools that enhanced the senses of a physician is a false analogy when compared with a tool that is intentionally designed to replace people: they are both the tool and the operator in one. That’s the big difference here. Drones aren’t putting a premium on chopper pilots, they’re putting them out of business. Before you take the next leap that I see a robot apocalypse approaching, that’s not what I predict. It will be much more sublime than that. We’ll boil our frog slowly in the water of job eliminations until we arrive at a recession, perhaps a depression. It’ll happen all over the world. Given how averse to the facts and easily swayed by conspiracies people have become, it may be difficult for a country like the USA to see what is happening. If we can see it, we won’t believe it. We’ll probably blame immigrants for stealing our jobs, political rivals and their values for destroying America, and so forth. There’ll be a lot of finger pointing and no action because businesses will be raking in record profits for a while. Corporations will promote the idea that it’s saving people money on products and services, and I’m certain that it will, at least until so many people are unemployed that they can’t afford them. They’ll also start by promoting how safe it is to have robotic/AI police, firemen, soldiers, drivers, and pilots, and they’ll be right. At some point however, there’ll be a tipping point with so many people losing their jobs. The old thought is that humans will just find other jobs, or that technology always creates jobs. We’ll assume that will be true because it superficially correlated in the past. Yet how will that model hold up against technology that is purposely designed to destroy jobs? What happens when we get to the point that no one wants to hire a human because AI can do everything better, cheaper, and faster? Who would hire a human? We’ll be antiques, just like horses. Why hire a human when AI is smarter? Why bother when a robot is stronger, faster, cheaper, makes fewer errors, doesn’t sleep, eat, sue, take maternity leave, require medical coverage? If a company can hire a robot to replace you, why wouldn’t it? The profit motive will steamroll all these ivory tower utopian scenarios in which AI and humans don’t compete in the marketplace. AI is coming for all jobs this time, not just boring, low-skilled, labor jobs. And it’ll be coming sooner rather than later. How far away are we from replacing the transportation industry with AI? A couple of years? Keep in mind that it’s the single largest sector of the economy in terms of employment. Also keep in mind that you only have to wipe out 25% of jobs to create a depression. Transportation losses will get us 20% of that pie in one shot.

  • I have been reading, cataloging and presenting about AI and healthcare (along with related topics) for several years now. This article tied together some of the more prominent innovations reliant on AI. However, I think AI will inevitably invade the human decision making process and our patients will be the better for it. Too often, the human error of medicine is borne of not incompetence, but simple error, bias, fatigue, overlooking a detail, etc. AI has none of those flaws. It never tires, has no bias, couldn’t care less if decades of conventional wisdom are simply not supported by the facts, it will produce answers that rely on statistical models and the culling of amounts of data that no one physical or PA can wrangle. I actually think that the future of medicine will be AI algorithms that will become the default decision makers and that the odd circumstance will be one in which the humans override those algorithms. Of course, the first time humans do that and the patient has a bad outcome, it will effectively end that practice for all providers. AI is not like the advances referenced in the article, the discovery and development of the stethoscope for example. AI is more akin to a single machine that can review, interpret, manage and distill data from the stethoscope, the patient’s entire record going back years, applicable demographic statistical data, the latest journal articles worldwide, and data points that we human medical providers are not even trained to look for. The eye image algorithmic device mentioned is a perfect example of this. There is no medical program that is teaching providers how to scan an image of a patient’s eye to reliably predict diabetic retinopathy risk. Why? Because humans cannot figure out what data in those images would raise or lower that risk. But, an AI algorithm after scanning more of such images than any one provider could reasonably review and retain, learned something about those images that we did not. It learned about patterns in those images which reliably predict a devastating diagnosis that if caught early can improve the outcomes for our patients. I do not see a future where this algorithm gets worse, but only better with more data. I do not see a future in which scientific studies will reveal that this algorithm becomes less accurate at these assessments and the humans figure out how to statistically out perform them. This is no new fangled stethoscope. AI is a machine that has never existed in medicine and it will change everything including becoming the decision making tool for most of medicine in the next ten years.

Comments are closed.