Two months ago, computer program AlphaGo stunned the world when it beat a human at Go, a popular Chinese board game, four times in five matches. Ahead of the first match, Lee Sedol, the human opponent and one of the best Go players in the world, confidently predicted winning all five matches. Until AlphaGo’s arrival, computer experts believed it would be incredibly difficult for a computer to beat a human at Go. However, since the surprising outcome in March, the Internet has been buzzing with speculations regarding the future of artificial intelligence, and how it might dominate human life, including healthcare.
The question of whether computers could replace doctors has been around for some time, but prevailing views have generally remained skeptical, mainly because of the complex and intuitive nature of patient care that goes well beyond rapid processing of data. After all, how could computers interpret qualitative signals and personalized nuances that patients reveal to their doctors?
However, a recent article that illustrated how a Toronto hospital went digital understandably gives us pause. It appears that technology and the power of artificial intelligence have already begun influencing patient care. And, a new Medicare proposal that requires doctors to be paid based on outcomes rather than volume could even accelerate how the healthcare industry incorporates the latest technologies into physicians’ daily practice. The outcome-based reimbursement policy will require physicians to incorporate the latest technology in managing patients’ data and treatment progress. Before you blink, a computer program like AlphaGo might be the doctor on your watch, telling you what disease you have.
So, what would be wrong with such a scenario of disruption? An incredibly powerful and omniscient computer program that can process trillions of instructions per second to deliver recommendations for patients is a very enticing thought. Without human error, healthcare can become simpler, more accessible, and more affordable- the very definition of disruption. But, before we get ahead of ourselves, there are several hurdles standing in the way, and they won’t likely be overcome in our lifetime.
First, diagnosing a disease is currently not a straightforward process. It begins with a general description of symptoms, followed by a series of tests that either lead to identification of the problem or a set of new tests. Unfortunately, almost all of the tests we utilize to narrow down the problem are still manual or semi-automated, requiring additional time and human involvement.
No matter how powerful and smart computers are, they will not be very effective at implementing manual tests to quickly identify the core of the problem. Imagine a person coming into a hospital with a fever. The speed at which the cause of the fever can be identified is limited to how quickly specific diagnostics can be performed. A super computer narrowing the possible cause down to a dozen issues, for example, does little to improve the speed. However, when faced with an excess of information and possibilities, humans have evolved to whittle down a large number of possible answers to the ones that seem the most important and relevant. In doing so, doctors are able to ask better questions, and arrive at an answer more quickly. Computers attempt to determine the cause by quickly verifying all possibilities, however they are limited by the speed of the diagnostic tests. Thus, until all of the key tests become completely automated and occur at or near the point of care, doctors will need to direct the technology. This evolution is likely to take a very long time.
Another hurdle is that the entire patient care continuum is a personalized “black box,” as there is little to no established process from one patient to the next. Every patient goes through a set of unique evaluations. In order for computers to replace humans, these personalized approaches will need to be replaced with standardized processes that can be optimized. While we advocate for more standardization and process-driven care models, these changes will likely be gradual.
Technology can help doctors be better, but it won’t replace them – at least not anytime soon. It is possible that some specialists could be replaced by primary care physicians, and some primary care doctors could be replaced by nurses, but the long-term outlook is that most doctors will continue to do their jobs, just better and faster with the help of technology.
Finally, the most difficult hurdles we see in machines replacing humans in healthcare is liability. One of the key debilitating cost burdens in healthcare is the cost of malpractice insurance imposed on physicians. The consequence of a single clinical mistake for a physician, especially if the mistake leads to patient death, is serious and permanent. A machine cannot take on such a burden, as there is no way of imposing exacting punishment on machines. Unless the paradigm of society were to change so that mistakes in healthcare were more benevolently viewed and treated, humans would not be able to give up total control to machines. Liability is also what will likely prevent non-physicians with super computers from replacing physicians. When a major mistake is made, everyone wants to blame the person with board certification.
Technology will be integral in changing how care is provided. But, just because technology can handle some of the routine care away from the hospital via sensors and wearable devices, that does not imply that care can be given in absence of physicians. Healthcare workers will remain integral to healthcare because the entire care system is dependent on their knowledge, experience, and intuition. Automating all of that will take a long time. Rather than look to computers to replace doctors, finding ways to push physicians and other healthcare professionals to deal with the most challenging problems, and leave routines tasks to the machines, is a much more achievable goal.