Even as big companies race to build up their artificial intelligence capabilities, the prospect of using A.I. technologies in healthcare raises more questions than answers.
Here are four key issues voiced by healthcare and technology experts in recent interviews with Xconomy:
—Will doctors, hospital administrators, and patients trust A.I. technologies? If the creators of a seemingly “black box” machine-learning system have trouble explaining how it makes decisions, will FDA regulators approve it?
—Will the tools slot in seamlessly to caregivers’ existing routines, or will the new technologies require a change in their work flows?
—What will the payment model look like for these technologies, particularly if they play a direct role in diagnosis and treatment?
—Will doctors and other caregivers lose jobs to A.I.? And if so, when?
On that last point, companies like IBM Watson Health and GE Healthcare insist that the A.I. technologies they’re developing are meant to assist caregivers, not replace them.
But investors and technologists interviewed by Xconomy say the automation threat is real, at least in some areas of healthcare. The most common idea seems to be that any job that involves a lot of image analysis—think radiology and dermatology—might see some threats in the coming years. That’s because advances in computer vision, pattern recognition, deep learning, and related technologies lend themselves well to such work.
It’s still early, but there have been studies showing medical image recognition software can be nearly as accurate as human experts. For example, a Stanford University study published in January in the journal Nature used deep neural networks to diagnose skin cancer from medical images. The software’s performance was on par with 21 dermatologists, the study’s authors wrote.
The person “who sits in a dark room reading images all day—that’s not a job I want to sign up for coming out of medical school,” says Michael Greeley, a partner with Boston-based healthcare technology investment firm Flare Capital Partners.
But software might not eliminate imaging-related doctor jobs entirely, Greeley says. The technology will likely reduce the number of images that are difficult to assess with certainty, but there will probably still be some that need a human eye. Those will go to the “grizzled radiologist” on staff, who has seen thousands of slides throughout his or her career, to make the final call. “The specialist will continue to have a role, but it will just be a more narrow role,” Greeley says.
Isaac Kohane, a doctor and the chair of Harvard Medical School’s department of biomedical informatics, predicts that A.I. will start having a tangible effect on medical imaging within three years. Pathologists and radiologists should start thinking about how they can adapt. “It’d be foolish not to think on how can I leverage that so I’m not put out of business,” he says.
Other potential applications of A.I. in healthcare are more “focal,” Kohane says, and maybe only cover 5 to 10 percent of a doctor’s daily routine. He cites examples like using software to determine if a patient’s heart failure is getting worse (and why), and suggesting which pairs of drugs are likely to cause side effects.
The possibility of implementing a “general-purpose A.I.” in healthcare is at least a decade away, Kohane estimates. He’s referring to a conversational software program that would assess symptoms reported by patients and deliver a “full diagnosis and treatment plan like a competent and well-educated primary care doctor.” Human doctors probably aren’t “sweating” the idea, he adds, at least for now.
Indeed, Alex Harding, a medical doctor conducting his residency at Massachusetts General Hospital, thinks A.I. in healthcare is “overhyped and not being used to a significant extent currently.”
“But I am a believer that artificial intelligence will become an important element in healthcare in the future,” says Harding, whose focus is primary care. “It’s years away, but I think that it’s something we’re going to have to become ready for and going to have to be thinking about.”
It sounds like he’s bracing for the possibility—not necessarily embracing it. In response to that suggestion, Harding says he’s unsure and wants to see how A.I. technologies get implemented in hospitals and clinics.
“I think it’s important not to take the human element out of healthcare,” he says.
He gives a real-life example of a patient with a chronic disease, who refuses to come into the hospital to have her laboratory tests checked. “I think for her it’s fear of having bad news,” Harding says. “No computer in the world is going to convince her to go in.” It takes a doctor to understand her and help make her comfortable enough to come in for her checkups, he says.
“Most of what we do in healthcare is not about making a diagnosis,” Harding says. “It’s about working with patients, who are humans, to find an appropriate treatment. That requires a human relationship … that develops over time and can’t be replaced by a computer.”