ARTIFICIAL INTELLIGENCE USED to mean something. Now, everything has AI. That app that delivers you late-night egg rolls? AI. The chatbot that pops up when you’re buying new kicks? AI. Tweets, stories, posts in your feed, the search results you return, even the people you swipe right or left; artificial intelligence had an invisible hand in what (and who) you see on the internet.
But in the walled-off world of health care, with its HIPAA laws and privacy hot buttons, AI is only just beginning to change the way doctors see, diagnose, treat, and monitor patients. The potential to save lives and money is tremendous; one report estimates big data-crunching algorithms could save medicine and pharma up to $100 billion a year, as a result of AI-assisted efficiencies in clinical trials, research, and decision-making in the doctor’s office. Which is why tech titans like IBM, Microsoft, Google, and Apple are spinning up their own AI health care pet projects. And why every health-focused startup pitching Silicon Valley VCs throws in a “machine learning” or “deep neural net” for good measure.
These algorithms get better the more data they see. And health data is practically hemorrhaging out of mobile devices, wearables, and electronic medical files. But their siloed storage systems don’t make it easy to share that data with each other, let alone with an artificial intelligence. Until that changes, AI won’t be curing the world of, well, probably anything.
Which is not to say AI in health care is all hype. Sure, Watson turned out to be less cancer-crushing computer prodigy and more very expensive electrical bill. But 2017 wasn’t all flops. In fact, this year saw artificial intelligence begin demonstrating real concrete usefulness inside exam rooms and out.
In the doctor’s office, AI is already helping dermatologists tell cancerous growths from harmless spots, diagnose rare genetic conditions using facial recognition algorithms, and lending an assist in reading X-rays and other medical images. Soon, it will be detecting signs of diabetes-related eye disease in India. But image classification isn’t the only thing it’s getting good at; AI can also mine text data. That kind of tech undergirds a platform that gives any primary care doc access to the expertise of specialists from all over the world. No more waiting six months for that referral you can’t really afford anyway. And after you get that diagnosis, you can now take home an AI-equipped robot to help you stick to your treatment plan. It nags, but it looks cute while it’s doing it.
Health care-focused AI has also seeped into virtual care, as medicine experiments with ways to offer preventive care and between-visit support via the omnipresent smartphone. Your phone no longer just tells you how to sleep better, eat healthier, exercise more, and keep a quiet mind. Now, AI can pick up patterns in the way you talk and text, to detect the first signs of depression and suicide risk. And it can help you deal with that stuff too. Amiable chatbots trained on cognitive behavioral therapy concepts are now helping people who can’t find time or money for a proper shrink. For veterans struggling with PTSD, researchers designed a human therapist avatar with a mind built by machine learning. Both approaches take advantage of the fact that people open up better to machines than other humans—the algorithms don’t judge.
And artificial intelligence is smartening up other devices, too. Deep neural software is making it easier to tune things like hearing aids and fancy new ultrasound machines. It’s making exoskeletons more responsive and artificial hands better at gripping (but not breaking) things.
Of course, as machine learning powers more and more medical device software, it’s made regulating them a whole lot trickier. This year the US Food and Drug Administration even had to create an entirely new digital health task force just to tackle it. How exactly do you regulate software that is always learning and evolving, constantly changing on the fly? What happens in a zero code world, where AI writes and rewrites its own instructions? Instead of trying to keep up with that radically different pace, the agency is piloting a new course—that certifies trusted companies with good track records, as opposed to individual software packages.
Still, those regulations will only control AI-informed devices, diagnostics, and treatments. The technology is seeping into the practice of medicine at every level, not just at the stage of final device approval. It’s now baked into the way biomedical researchers sift through tsunamis of geneticdata and pharma firms discover new drugs. It’s how public health officials predict the next epidemic, and keep track of opioid hot spots. And it’s increasingly how doctors and scientists try to make sense their data-drenched realities. As AI opens those new avenues of understanding and treating human disease, it’s important to remember that algorithms, like people, are imperfect. They’re only as good as the data they see and the biases they carry.
No matter how many black box neural networks start finding their way into the health care system, medicine is still fundamentally a human endeavor. And people don’t always do what’s best for them, even on a doctor’s orders. Which means the biggest challenge in health care isn’t about changing people’s bodies, but about changing people’s minds. And that’s not the kind of intelligence computers are good at. AI won’t be replacing MDs, anytime soon. But it is coming for their fax machines.