Health care has been slow to adopt Artificial Intelligence (AI), but we are on the cusp of astonishing change
In the fall of 2016, Geoffrey Hinton, widely considered the godfather of Artificial Intelligence (AI), told a conference that medical schools should stop training radiologists now, because it was “completely obvious” that within five years machines were going to be better than humans.
Fast forward eight later, medical schools are still training radiologists who continue to play a critical role in the health system. Although we are more than a decade into the AI revolution that began with deep learning, the kind of changes – both good and bad – many had predicted have not yet materialized.
Delays in wide-scale implementation are an essential part of any technology with the power to reshape society. It was clear, for example, in the 1880s that electricity was going to be revolutionary, but it wasn’t until the 1920s that most households and factories were electrified.
And in the context of AI, we are standing a lot closer to 1880 than we are to 1920, said Dr. Avi Goldfarb, PhD, the Rotman Chair in Artificial Intelligence and Healthcare at the Rotman School of Management, University of Toronto. Dr. Goldfarb spoke recently at the Family Medicine Summit, a conference organized by the Ontario College of Family Physicians.
But even relative to other sectors, health care is a “laggard” in its adoption of AI. Some of that might be explained by the regulatory approval needed in health care, which might mute enthusiasm to explore new opportunities. But it’s just as likely that slow adoption is a result of the AI experience being, to date, a little underwhelming in health care.
“Maybe, in radiology, it’s a little better at detecting an anomaly. Maybe it can detect sepsis a little bit earlier, some of the time. Because it is only allowing you to do what you’ve always done but just a little bit better, it’s not providing a lot of incentive,” he said.
In his book, Power and Prediction: The Disruptive Economics of Artificial Intelligence (co-authored with Ajay Agrawal and Joshua Gans) Dr. Goldfarb refers to the present as “The Between Times,” the period between envisioning what is possible with AI and its widespread adoption and optimization.
To move beyond The Between Times, we will need to evolve beyond the “point solutions” focused on specific tasks employed in the existing health-care system to “system solutions” that create transformational change, he said.
It’s just as likely that slow adoption is a result of the AI experience being, to date, a little underwhelming in health care.
For AI to fully take flight, he says, we must imagine how people’s health can be best served in a freshly designed system that has access to new prediction technology. That means rethinking medical school training, delivery procedures, compensation, privacy and liability. It means, he said, adopting a system mindset.
For example, Dr. Goldfarb describes how AI, in full throttle, could redesign the care received in an emergency department. Imagine, he says, AI can predict that a patient arriving at the ED is having a heart attack with extraordinary accuracy. Ultimately, the doctors and administrators could agree that predictions are so accurate that it makes sense to skip stress tests entirely and go straight to catheterization.
“In fact, with enough data, it might be possible to move those predictions out of the triage space in a hospital’s emergency department all together and back into a patient’s home. Thus, before the ambulance is called, it is possible to imagine high-quality predictions that provide a diagnosis,” he wrote in his book.
In such instances, would the patient even need to come into the emergency department at all? Perhaps they could get sent to the relevant medical department instead. Or if the patient is diagnosed with a condition a primary care physician can help treat, then the patient would not need to even step foot in the hospital.
It all begins with identifying the key decisions, speculating on what is possible if the predictions become highly accurate and then “reimagining the types of systems that can exploit these predictions in a manner optimized for mission success.”
The process of discovery takes time, he says, as people evolve in their understanding of what is possible. After all, nobody living in 1880 could have described the systems borne of electricity, reminds Dr. Goldfarb.
