The associate chief of the MS division and professor of neurology at the University of Pennsylvania described the relationship between artificial intelligence and medicine, and how he sees it evolving in the future.
Joseph Berger, MD, associate chief of the MS division and professor of neurology at Penn Medicine
Joseph Berger, MD
The 2019 American Academy of Neurology Annual Meeting in Philadelphia hosted a debate on whether a robot with AI could replace the role of care provided by neurologists.
Joseph Berger, MD, associate chief of the MS division and professor of neurology at the University of Pennsylvania, spoke with NeurologyLive to share his thoughts on this topic and discuss the uniqueness within neurology that would make the transition to AI different from other specializations.
Berger explained that the investment to replace neurologists, because of the complexity of what they do, how difficult it would be, how few of them there are, and how little they contribute to the cost of medical care, would not make it worthwhile, it's simply not worth the investment.
Joseph Berger, MD: This year the American Academy of Neurology introduced something very interesting into the program, which is a debate that is on the topic of whether robots and artificial intelligence can replace neurologists. It's not something inconceivable because there are physicians that ultimately will be replaced by artificial intelligence, those people that are dependent on pattern recognition.
We're seeing it now in such disciplines as radiology, where lesions on chest x-rays are picked up better by artificial intelligence than people reading them, dermatology, another form of pattern recognition, the funduscopic examination, electrocardiographic interpretation, probably at some point in time electroencephalographic interpretation, those things will be replaced—I mean there's no reason to have a physician reading over them, maybe for those that are outliers that look very unusual, it’s necessary to have a physician look at them but otherwise the machine could do it as well, if not better, than any human being.
There's something unique about what neurologists do and a number of years ago in JAMA Neurology I wrote a paper entitled Neurologists the Last Physician Scientist and in it what I address is that we are the last physicians that are still performing detailed physical examinations and there's a reason that we do that, and that is that when we hear someone's story, we generate hypotheses—just like a scientist generates hypotheses based on observations and those hypotheses are importantly where a lesion is located in the central nervous system or the peripheral nervous system and we test that hypotheses by actually examining the patient. The examination is a critical component of what we do as neurologists and it'd be extraordinarily difficult to program a robot to do the types of examinations that we do, which we do with tools that are essentially primitive, a reflex hammer, a pin, a tuning fork, these are what we do our examinations with.
In the course of the examination not only are we testing the hypothesis of where the lesion is but at one in the same time, we're developing a bond with the patient that doesn't exist when that component of the examination isn't done. In fact, what we've seen, so if you look in 1975 the average time that a physician allotted for a new patient, this is physician in general, not just neurologist, was one hour the average amount of time in 1975 that a physician allowed it for a follow-up patient was 30 minutes, today the average time for a new patient is 12 minutes and the average time for a follow-up patient is 7 minutes, you cannot develop the bond in that period of time. There were people back in the early part of the 20th century, in the late part of the 19th century, including William Osler very famous neurologist who said ‘that you cannot adequately evaluate a patient unless you allot at least one half hour, and they don't feel satisfied unless you do,’ and that's true today as well.
What happens when you walk into a physician's office these days? They've got their nose in the computer, they're busily typing away, they don't even look at you, and they hardly examine you, it's a little different for the neurologist, he may have his nose in the computer and I would argue that that's not how an examination should be performed, but they do examine you, a good neurologist examines you and they examine you for the very reasons that I've stated. I think that neurologists are quite unique and importantly if you were to ask individuals who develop AI when various people in the workforce are going to be replaced, they'll tell you that the person selling retail may be replaced in 12 years, that even the surgeon may be replaced in 35, and they themselves may be replaced in 125 years, there was actually a study that queried individuals that design artificial intelligence and these are the numbers that they came up with in the aggregate.
I would say that the investment to replace neurologists, because of the complexity of what they do, how difficult it would be, how few of them there are, and how little they contribute to the cost of medical care, would not make it worthwhile, it's simply not worth the investment.
There are physicians who I think are going to be replaced ultimately by artificial intelligence, or largely replaced by artificial intelligence. Very often the diseases that we deal with require compassion and empathy, I mean can you imagine a robot telling somebody that they had amyotrophic lateral sclerosis or that they had Alzheimer disease or that they had Creutzfeldt-Jakob Disease and we're destined to die or telling the family. You also have to gauge the individual, what is it that they want to know, you have to get a sense because there's nothing formulaic about it, it's something that you learn over the course of years on how to convey that information.
I will say that artificial intelligence will augment what we do, there's no question about it, it's doing that now. In my mind there is a role for artificial intelligence when it comes to the practice of neurology. What is that role? That role is it's going to help us in terms of diagnosis, I mean it may suggest when you put a diagnosis in the record, it may suggest others that may make more sense, it will probably suggest some testing that would be worthwhile that you may or may not have thought of, it might help you in terms of interpreting that testing, saying this test has a certain sensitivity and specificity that in this setting means that it's not particularly valuable, and as we have increasing numbers of therapies for a wide variety of conditions, it's going to perhaps be able to search the most recent medical literature and say to you, when you put that diagnosis in and you select a therapy, it may say you know what this might be a better therapy given this patient's demographics, etc. I think it's going to be very helpful in terms of assisting with the diagnosis, not necessarily making the diagnosis, I think it's going to be very helpful in terms of helping you select the appropriate tests, interpreting those tests, and then suggesting therapies, but it's not going to replace the neurologists for the uniqueness of what the neurologist does.
When you obtain a history as a physician, there's a certain amount of interpretation that goes on, so not everybody knows the medical terminology, sometimes even the lay term analogy is mistaken. Particularly when I was practicing elsewhere not so much in Philadelphia people would come in and they say I have numbness, I'm numb or my left side, and you say well what do you mean by numb, and it turns out they didn't mean numbness, they didn't mean a loss of sensation, they meant weakness. Well if a computer or robot was to get numbness, what they would think would be very different than if they get weakness. By the same token, if a patient comes in and says doc, I'm dizzy, what does that mean? Are you dizzy because you're lightheaded and you feel like you're about to faint or are you dizzy because you feel like you just stepped off a merry-go-round and you're spinning around—very, very different things. There's a mound of interpretation in the language that's very important and would, in my mind, be very difficult to program a robot to do. I think that that's another important component and that goes to the history taking before the physical examination.
Transcript edited for clarity.