While generative artificial intelligence has potential for healthcare, its actual uses are "overhyped," according to B.J. Moore, CIO of Renton, Wash.-based Providence.
"You would think it's already been used everywhere, based upon everybody talking about it," he told Becker's. "The capabilities are properly hyped, but how far along we are and how people are leveraging it today is overhyped."
His 51-hospital system is putting it to use, albeit in "baby steps," he said. Providence started piloting generative AI for inbox management of physician emails in May, cutting the response time from three days to one. Providence customized the development of that platform using a large language model from Microsoft.
"It'll say, 'This message is about a follow-up appointment, this one is about pain management, this one is about symptoms from medications,'" Mr. Moore explained.
Generative AI has also been incorporated into the Dragon Ambient eXperience, or DAX, clinical documentation program that Providence uses from Microsoft subsidiary Nuance. Mr. Moore said it used to take three or four hours for a human being to review a note and upload it to Epic; now it posts to the EHR in near real time.
"About six months ago, we did a survey of our physicians of what their response would be if we took DAX away from them, and 85 percent said it would be very unfavorable," he said. "It was a good product a year ago. It is only a better product today."
In addition, Providence is using generative AI for a physician education and referral platform called MedPearl and for its AI chatbot, Grace. Providence is also the largest healthcare provider on Oracle Cloud and is partnering with the software company on generative AI solutions, Mr. Moore said.
For generative AI to reach its potential in healthcare, tech leaders in the industry will need to take it slow and make sure to get things right, Mr. Moore said, using it as a "co-pilot" for physicians and nurses. He cited IBM's Watson as an example of how it could go wrong.
"One of the reasons Watson failed is they went immediately to trying to cure cancer, and I don't think we can take that kind of approach," he said. "So we help you manage your inbox and then help you do patient reminders then help make that diagnostic recommendation and recommendation on imaging. It's a continuum, baby steps."