AI will outperform the best medical experts

Q. Let’s start with a big-picture question: How do you foresee AI impacting patient care? As we use data and analytics to shift towards prediction and prevention – and away from costly post-illness treatment – one would hope that AI will prevent many people from becoming patients in the first place. Does such a positive trend seem likely to you?

A. Artificial Intelligence (AI) will impact patient care in many different ways – but probably the most dramatic will be that simple, low-cost home equipment will continuously monitor our health, and notify our medical care professionals (and us!) of any potential issues or problems, often before we experience any symptoms or any damage is done.

This will turn our relationship with health care upside-down. Today, you (the patient) are responsible for monitoring your own medical condition. If and when you decide you may have a problem or some issue becomes sufficiently bothersome, you must take the initiative to contact your doctor. In the future, in part due to advances in AI, many common medical problems will be identified by equipment in your home or workplace, and these devices will transmit relevant diagnostic information directly to your doctor (or you) for analysis. This equipment may be worn (such as your watch) or environmental (such as an air quality monitor).

This will allow you to address or resolve medical issues such as high blood pressure, or low blood oxygen, before they become problems. It will also save enormous amounts of time and money currently wasted on unnecessary visits to doctor’s offices for test results that turn out to be medically insignificant.

AI comes into this picture in two ways. First, much of the required computation will be done “locally,” meaning on a home medical device, or on your phone. For instance, a specialized phone application may allow you to take a picture of a skin lesion and diagnose whether it is a dangerous cancer just as accurately as a dermatologist would do.

Second, some devices may transmit information to a medical monitoring facility that can combine your information with the information of others to act as an early warning system for infectious diseases.

For instance, it may be possible simply by sampling air in public places to determine whether a variant of Covid is present in the environment. By “connecting the dots” in different locations, such a system might be able to identify specific asymptomatic individuals who are developing and spreading the infection.

This, of course, raises many questions of privacy, which will have to be addressed. But it will dramatically improve overall health outcomes and reduce the cost of care.

Q. To what extent will AI and machine learning improve the practice of diagnosing diseases and medical conditions? A 2014 study by Johns Hopkins said misdiagnosis was the leading cause of “serious medical errors” in the United States, and that between 40,000 and 80,000 deaths linked to misdiagnosis occur in US hospitals each year. Should we be looking forward to a golden age of medical diagnosis, where diseases are caught and treated earlier?

A. In the future, diagnosis by some form of AI or assistance to medical professionals will certainly improve the accuracy of diagnoses and avoid all manner of medical errors. Some fields, such as radiology, are purely based on data (as opposed to a physical examination of a patient), and so are subject to types of analysis at which modern AI systems excel. But new AI-based systems may go much further than the equipment now commonly used to monitor patients in hospital rooms and intensive care facilities.

For instance, analysis of images streaming from cameras may be able to identify ambulatory patients with balance problems or other neurological conditions. Or, cameras monitoring a surgical procedure taking place in an operating room may be able to spot potentially contaminating conditions, such as an instrument that was inadvertently placed on a non-sterile surface, or used improperly. And in the future, AI may be able to identify a variety of conditions, both physical and mental, by just analyzing the sound of your voice.

AI brings a new class of tools to diagnostic medicine. Because these systems can take into account not only the latest research, but also the knowledge and experience of large numbers of practitioners, they can often outperform even the best medical experts, perform more consistently, and bring quality health care to populations now underserved by the medical system.

Q. What about the use of machine learning for drug discovery? Will we get better drugs, faster, thanks to AI and machine learning?

A. The simple answer is “yes.” Just last year, the German biotechnology company Evotec began testing a new anti-cancer molecule identified by an AI program that identifies promising small-molecule drugs. Previously, the traditional discovery process might have taken four or five years to come up with this candidate, but the company’s AI technology reduced the time required to just eight months.

This is not an isolated example, of course. Exploration for new drugs, aided or done entirely by AI, is advancing in a wide variety of fields, targeting such intransigent conditions as lupus, glioblastoma, aggressive cancers, and fibrotic diseases, to name just a few.

Q. Assuming that AI will bring about improvements in these and other areas of medicine, how soon do you think those changes will come about? 

A. As the expression goes, “It’s difficult to make predictions, especially about the future.” Many of these changes are happening right now, but with recent advances in AI and medicine, it’s reasonable to expect that the pace of change will increase notably in the next few years and decades.

Probably most of the effects of this technology will be invisible to the average patient. They will impact aspects of medical care that take place behind-the-scenes, improving diagnosis, speeding treatments, eliminating waste, and making quality care more widely available.

But some changes will take place right under our noses – sometimes literally. As you may know, some dogs are able to detect cancers in their owners by their smell. Advanced electronic devices are under development to detect a variety of medical conditions merely by sampling our breath or bodily fluids. Just as epidemiologists now sample sewage to detect outbreaks of contagious diseases, a slimmed-down version of the same technology may be a standard feature of future toilets!

Q. How might AI change the relationship between the patient and the doctor? As robotics and AI enter the picture, will the emotional element of caregiving be the only thing left that humans can provide – and if so, will we really need a highly trained, highly paid doctor to offer it?

A. The human touch, in many fields, is overrated, in my opinion. The purpose of medicine isn’t to comfort you emotionally (though that’s an important aspect of care), it’s to identify medical problems and address them. I don’t miss the days when I had to speak to a human teller to withdraw cash from my bank account, or wait for some hapless attendant to pump gas into my car; I’d rather have a computer address my medical problem in the middle of the night than wait for my doctor to wake up and return my call.

That said, there will always be a need for human doctors, at least for the foreseeable future. We aren’t close to having computer-based interviewers that have the breadth of experience and judgment to assess how much pain you are in, or suspect that you are hiding some important detail due to embarrassment. Sometimes, the doctor’s orders may have more impact than a computer printout in goading you into more healthy behaviors.

Q. In your book, Humans Need Not Apply, you talk about the potential disruptive effects that automation could have on employment. How do you think AI might affect the labor market in the field of healthcare?

A. History shows that automation doesn’t so much replace people as change the nature of work. Tellers who used to spend time in routine transactions like taking deposits and handing out cash have been replaced by loan officers and account managers who help bank customers with more complex tasks. There are a lot more auto mechanics today than there were “stablemen” who used to care for the horses we used for transportation, centuries ago. Legions of artists who used to draw individual “cels” for cartoons have been replaced by much larger armies of graphic designers making CGI (computer-generated imagery) for films.

Somehow, despite centuries of advances in automation, here we are at full employment, with a much larger population. So I don’t think doctors, nurses, or other medical professionals are in any significant danger of joining the unemployment line any time soon – although the nature of their jobs is likely to evolve significantly.

Q. The expectation is that AI’s predictive abilities will persuade people to change their behavior in healthful ways. But might it have the opposite effect? If people are told they are at risk for some acute illness, or dementia, might they act even more recklessly, reasoning that they have little to lose and might as well enjoy the time they have left?

A. You make a very good point. There’s considerable evidence that as we make things safer, people feel free to act more dangerously. Nowhere is this clearer than with improvements in automotive safety. With collision detectors, and features such as back-up and lane-departure warnings, there’s no question that drivers pay less attention to safety while executing these maneuvers. And why not? We shouldn’t feel ashamed that machines can make us better drivers and relieve us of effort.

The same effect is likely to be true of medical problems. If your watch or phone can tell you when you’ve had too much to drink, why bother to keep track yourself? If some new technology – whether based on AI or not – offers a cure for your hangover, why not imbibe at will? If your computer warns you to get up and move around every once in a while, you don’t need to concern yourself with how sedentary you are. The technology simply provides more options; how you choose to use that newfound freedom is a personal choice.

Q. Looking ahead somewhere between five and, say, 20 years, what excites you about the future of AI and healthcare? What, if anything, worries you – and why?

A. Being 70 years old, I tend to be more focused on the short term. Frankly, I don’t think much will change in the next five years – or at least it won’t feel like it. Longer term, there will definitely be substantial progress in diagnosing medical problems, with greater accuracy, at lower cost, and with new and more effective treatments.

But I don’t think you will wake up someday to a new world of medical care; the progress will seem gradual and incremental, until you realize that your friends are living much longer than your parents’ friends did, and your quality of life is far better than that of previous generations of people your age.

Despite that rosy prediction, there are definitely issues that will require considerable social capital and political will to address. One is that when it comes to medicine, one size does not fit all. Different races and genetic groups have different characteristics, but unless these groups are large, there will be considerable economic pressure to streamline diagnosis and treatment to address only the largest and therefore most profitable groups of people.

Also, the privacy implications of having your bodily functions monitored, either at home or in public, are concerning. Balancing your right to behave as you wish against society’s interest in delivering improved health outcomes will be subject to bias, inequality, and mismanagement.