How to Integrate AI Into Workflows

An AI tool can be developed and shown to work under some kind of clinical testing, ideal conditions in a way. But when you actually implement it into the real world, then maybe it just is the form as well. And maybe the performance gets worse over time. We think about it a lot with AI because we're training these models and then we validate these models. So one of the challenges that we can come back to is this whole concept of where does the data come from? Right?