Renton, Wash.-based Providence didn't want to ban ChatGPT and force employees to use it on their personal computers, unmonitored. So the health system created its own version.
The 52-hospital system rolled out ProvidenceChat to its 120,000 staff members Dec. 1, allowing them to privately use generative artificial intelligence and not worry about the data being shared online or with tech companies.
"I've heard a lot of health systems say we're going to block OpenAI and ChatGPT on our networks," Providence CIO B.J. Moore told Becker's. "My approach is the genie's out of the bottle. People are going to use it regardless. So how do you proactively build safeguards in place to allow them to do that?"
So last summer, his team started building a private version of the platform that doesn't learn from users' prompts or send the data back to Microsoft or OpenAI. Providence can also monitor the questions employees are asking (the system is transparent that it is recording you).
"By having ProvidenceChat, we can see how people are using it, to not only make sure people are using it for good, but then to help as we build solutions," Mr. Moore said.
If people are repeatedly asking it the same question, maybe that's a program the health system can implement in its Epic EHR, he explained.
The most popular inquiries so far have been related to IT and coding (32.3%), writing assistance (18.8%), general (17.5%), proprietary data (10.4%), and public knowledge about Providence (4.72%), according to a health system spokesperson. The remaining percentage includes general medical questions, data mapping, analysis, management, and legal and compliance issues.
Some common questions have included: "How do I use the coding system for finance operations?" "Can you help me draft an email to a client?" "Can you provide a recipe for a healthy meal?" "Can you suggest some exercises for a quick break?" and "Can you provide the latest data on the predictive payroll effort?"
As of mid-January, 3,400 staffers have accessed ProvidenceChat, asking a total of 82,000 questions. The majority work in IT (46%) followed by human resources (20%) and clinical support (14%).
Employees can pose healthcare-related queries — Providence just requests they don't make clinical decisions based on the answers, Mr. Moore said. So far, those types of inquiries have only made up a small percentage of the total.
"We want them to ask it clinical questions so people understand the capabilities and what works and doesn't work. So they can understand the power of it," Mr. Moore said. "And then my team on the backend can see — if we were to use it for clinical areas — the prompts that people are asking. So then it allows us to go, 'Oh, people are asking pretty consistently about blood test results. That's a capability we should work with Epic to make sure is in the Epic product.'"