ChatGPT as a Teammate

My name is Cornelius James, I'm an assistant professor in the Department of Internal Medicine, Pediatrics and Learning Health Sciences here at the University of Michigan. And I lead a team called the data augmented technology assistant medical decision making team or DataMD. And what we do is develop curricula that are intended to teach learners across the continuum of medical education about artificial intelligence and machine learning. There will essentially be a continuum when it comes to how individuals interact with artificial intelligence and machine learning. So, quite often, you'll hear it referred to as a tool which in some cases will be appropriate, and in other cases it will be a teammate. So if you kind of think about a tool, a tool it typically doesn't interact with you, right? Or a tool does not, in many respects, engage with you, right? So there will be some instances in which artificial intelligence will not have much engagement and or interaction, direct interaction with an individual, a clinician, for example. Whereas in other situations, the artificial intelligence could have a bidirectional exchange with either a patient, a medical student, a clinician, I think, for example, of chatbots. So chatbots are tools where people can have it on their phone, and individuals can literally have conversations back and forth with that technology. So in that respect, I would consider that to potentially be a teammate given the bidirectional exchange that's happening, as opposed to other technologies, again, where it's simply entering something into a computer and then you get something that's going to be an output that will come back to you. But in many cases there's not as much of that bidirectional exchange. And I'll add to that, AI is going to have certain levels or differing levels of autonomy, right? So there will be some instances in which the artificial intelligence will not require any direct human input or response. And there will be other times where humans will oversee everything that the AI does prior to the AI making a response, or giving a response, or being able to act on whatever it received as an input. So you can imagine a tool typically does not, we don't talk about a tool generally having a lot of autonomy, right? Whereas with artificial intelligence, you're going to have either AI that is not autonomous at all, and it's all directed and driven by a human. And then you'll have AI that will be semi autonomous, where they'll kind of work together as a team to perform certain tasks. And then you'll have AI that is completely autonomous, where it doesn't rely on any input or engagement from a clinician to make a decision.