BOSTON – The excitement around artificial intelligence is genuine. The promise and potential of what it can do for healthcare is very real. But there’s still a lot of unrealistic expectation of what AI and machine learning will mean for care delivery, said Dr. Anthony Chang.
Chang, chief intelligence and innovation officer at Children’s Hospital and Orange County, delivered the Day 2 keynote address at the HIMSS Machine Learning & AI for Healthcare event here in Boston.
There’s already a lot going on with the technology in hospitals, as evidenced by the numerous case studies presented here these past two days. But Chang said the best is yet to come for AI in healthcare.
He pointed to Amara’s Law: “We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.”
When it comes to AI and machine learning, many in healthcare are “anxious … waiting for a lot of delivery right now,” said Chang. “But if we can just hang in for a few years,” the benefits will be immense.
For instance, he sees AI and ML paving the way toward “true precision and personalized medicine in the coming decades.”
But in the shorter term, there are still some misconceptions that need clearing up, he said. Here are 10 of them:
Clinicians will be replaced by AI
“No clinician has ever said that we don’t need cardiologists and radiologists,” said Chang. Despite the fears that image reading will be the exclusive province of computer algorithms in the years ahead.
“Clinicians will have to teach the machines [cognition] in the coming decade,” he said. “We’re a long way from robots having to do a complicated procedure.”
He added: “I predict today that there are going to be more radiologists in the future, not less, because AI will make their jobs more interesting.”
AI needs to be added to every aspect of healthcare
“Not every part of the hospital needs deep learning,” said Chang. “Some will benefit from easy, simple algorithms.”
AI conquered Go, so biomedicine is easy
Healthcare is not the same as an ancient Chinese board game, as complex and abstract as that game may be.
In medicine, “there are multiple things happening at once, most of the time I don’t have all the data,” said Chang. “It’s much more like the real-time strategy games.”
That said, however, “I think the day will come when a machine intelligent algorithm can at least partly run an ICU,” he predicted.
Deep learning will be the main AI tool for a long time
“Deep learning is exciting but should only be the beginning,” said Chang. “There’s more excitement to come in the next decades.”
A common way of framing humans’ relations with AI, he explained, is that “machines are fast but kind of stupid; humans are slow but kind of smart.”
The future, he said, will be one where computers are both fast and smart, and today’s deep learning will be supplanted by a much more cognitive technology.
You have to program to contribute
“You don’t have to be a programmer to make a contribution” to the beneficial evolution of AI, said Chang. “Clinicians and other healthcare professionals can really help by looking at the foundational layer,” helping build that progression from data to information to knowledge to intelligence, he said.
“Hopefully, one day the wisdom part will be really big because everyone is contributing,” said Chang.
We need more data for deep learning in medicine
“Deep learning is data hungry,” said Chang. The challenge, however, is that not every clinical use case has the data to feed it.
“I don’t have 100,000 MRIs of a rare disease to give everyone,” he explained.
That’s why, more and more, “data scientists are learning to deal with small data, or little data,” he said. They’re getting better at synthesizing and testing with their own data, and soon “deep learning will not require lots of data.”
AI isn’t made for primary care
“Going for the quadruple aim takes a lot of work,” said Chang. From chart review and documentation help for the PCP, to digital health coaching on the patient side, AI and machine learning have a big role to play in primary care and population health management.
AI will make clinicians less human
“Think of AI in medicine as your clinical GPS,” said Chang. “It’s not going to get it right 100 percent of the time. Humans absolutely need to be engaged, and have oversight.”
AI is a black box
“If you don’t understand it, it will stay a black box,” said Chang. “Clinicians need to understand the risk-benefit ratio for every tool.”
Some clinicians still resent the inscrutable nature of AI, of course, but as it continues to evolve and show its worth, that skepticism may fade, he said. “If things are mostly benefit but little downside, you’re going to be more forgiving of autonomy.”
AI in medicine will be here in the future
AI and machine learning are here now, of course. And they will get better. But that improved future state needs guidance too.
“We all have to create that future,” said Chang, and that will require “clinicians working much closer with data scientists.”
The key is to “stay patient,” he said. “Don’t overhype, but don’t underestimate what it can do 20 years from now.”