AI Can Help in Healthcare, but Normal Intelligence Is Still Needed, Panelists Say at JP Morgan Conference

Published on: 

Artificial intelligence (AI) can be trained with data sets to recognize patterns and improve healthcare outcomes. But first, healthcare needs to get better at using normal intelligence to solve problems, said panelists at the JP Morgan Healthcare Conference in San Francisco, California.

Artificial intelligence (AI) can be trained with data sets to recognize patterns and improve healthcare outcomes. However, healthcare might not be ready for AI—it sometimes isn’t even good at using normal intelligence, said Andy Slavitt, former acting administrator of CMS, during a panel discussion at the JP Morgan Healthcare Conference in San Francisco, California.

There is plenty of promise in AI, but it is not a solution in healthcare, it is a tool that will allow providers and researchers to address complicated problems more effectively, said Jamie Heywood, founder and chairman of PatientsLikeMe, expressing a sentiment that the rest of the panel largely agreed with.

“It’s a little bit of the best-of-times worst-of-times story,” Slavitt said.

We’re at a moment where we can see the next generation having their genomics installed on a liver that was created with a 3D printer, but healthcare has the problem of chasing new ideas and putting them on the backs of old ones, which adds a layer of complexity instead of simplifying it, he explained.


For example, providers know there are people in their community that show up at the emergency department with enormous cost that can be reduced or eliminated entirely simply by reaching out and delivering their medications.

“That doesn’t take AI,” Slavitt said. “That takes a fundamental ability to ask questions, to sort through what’s going on in our community, and put in place effective processes and programs. We’re not even at that level yet.”

Gini Deshande, CEO of NuMedii, explained that AI is useful in drug discovery. One of the biggest challenges in drug discovery has been a lack of understanding about the complexity of a disease. AI, coupled with good data, can begin to look for subsets of patients, identify them earlier, and identify therapeutics that will help those patients effectively.

She views AI as being able to remove some of the trial and error components of drug discovery and systematize the process better.

“We’ve come to appreciate what’s happening in conditions like oncology—it’s become very clear that 1 disease is not 1 disease,” she said. “It’s a subset of many diseases and there’s an enormous amount of data that’s now being collected.” AI will help make sense of that data.

AI can also help with some of the human components of healthcare. As a physician, Jordan Shlain, MD, founder of HealthLoop, recognizes that it can be difficult for a provider to know about patients in their lives when they’re not at the doctor’s office. That information can be collected via AI and packaged out to be actionable.

The human interaction in healthcare will always remain important. Shlain explained that as a primary care provider, he doesn’t see AI replacing him any time soon.

“I’ll be kind of the last one to go, and the guy after me will be the priest and the rabbi,” he joked. “And that’s when the AI will actually take over the universe.”

Slavitt closed out the panel, echoing the importance of getting the human component and normal intelligence right before AI can improve healthcare. There remains an element of simply working hard to solve challenges in healthcare that only require a basic level of intelligence, he said.

“We have these sort of really interesting capabilities, and I think we would all agree we’d rather pay for 50 vans to drive around San Francisco than the cost of a bunch of people showing up in the emergency [department] because they didn’t get their medication and didn’t get to their appointments,” Slavitt said.