The type of robot Shah used. SoftBank
At least that’s what Julie Shah — an MIT professor who leads the Interactive Robotics Group at the university’s Computer Science and Artificial Intelligence Laboratory — is working on.
Shah recently programmed a Nao robot to watch how medical assistants perform tasks so it could help assign patients to the appropriate doctor and move patients to operating rooms and beds. That may seem like a simple task, but can actually be quite complex to handle.
At Beth Israel Deaconess Medical Center, the head nurse of the labor ward coordinates 10 nurses, 20 patients, and 20 rooms at the same time. That adds up to two to the one millionth power scheduling possibilities — not something easy to code.
“The hospital domain is really interesting because it’s more complex than an air traffic controller’s domain,
Shah told Tech Insider. “We can’t compute an optimal solution to the problem they’re solving, yet magically some experts are able to do this after years of training quite well.”
“The question was, then, can we reproduce their performance to have robots train novice nurses and doctors?” she continued.
After observing how nurses made decisions, the robot tried to coordinate some tasks by itself. The Nao robot did extremely well — nurses took the bot’s recommendations 90% of the time.
“The machine can potentially train a novice when an actual human expert is not available,” Shah said. “The ultimate goal would be superhuman performance through collaboration.”
So when can we see robots training medical personnel?
“10 years is definitely within the realm of taking something from the lab and scaling it,” she said.