Is AI as Smart as It Thinks It Is?

Ryan Black
AUGUST 08, 2017

“The aim of medicine is to prevent disease and prolong life; the ideal of medicine is to eliminate the need of a physician.”

William James Mayo penned the above in 1928. Though Mayo didn’t have AI in mind – it would be another 27 years before the term’s inception – the statement serves as a reminder that the noisy hype surrounding AI may be drowning out reality.

“There’s a lot more buzz than substance, that’s exactly what I would call it right now,” Dr. Jay Anders, CMO of Medicomp Systems, told Healthcare Analytics News. “I think AI in healthcare is struggling to figure out exactly what its place is. It went, I would say, almost off the rails.”

Anders believes that the infatuation with the concept of AI has outstripped its real potential and muddied the waters. A massive sticking point conveyed by experts to Healthcare Analytics News is a lack of concrete definition for the term "artificial intelligence" as it relates to healthcare.

“AI is unfortunately an overloaded, overused, and frequently misunderstood term,” said Anand Shroff, the CTO of Health Fidelity. “To me, artificial intelligence is the ability to help physicians with information that they may not necessarily have handy, by leveraging massive data sources and by cross-searching different sets of information to provide lots of valuable input to help make the best possible decisions.”

Explaining the analytics work of his company, Clarify Health, CEO Jean Drouin pointed out that his firm has tried to avoid the term. The company brings together wide swaths of disparate patient and demographic data to try to create best-course recommendations for physicians.

“A lot of companies might call that AI, but we’ve been careful about using the term because it’s so overhyped,” he said. “This was a terrific point in the Stanford article, I absolutely agree that you have to be very careful not to over-focus on the idea that [we can predict] a patient is going to have an event on this date. We are a long, long way away from that.”

The Stanford article he referenced was published in June by the New England Journal of Medicine. Two researchers, Jonathan H. Chen, MD, PhD, and Steven M. Asch, MD, MPH, both of Stanford, struck a more measured tone on AI’s role in medicine. The piece in no way overlooks the potential of big data, AI, and machine learning for medicine, but lists a series of sticking points that often go unmentioned.

“Whether such artificial-intelligence systems are ‘smarter’ than human practitioners makes for a stimulating debate—but is largely irrelevant,” they wrote.

Their aim, they indicate, is to encourage the use of predictive analytics in proven ways for better care. They posit that medicine is “at the peak of inflated expectations” and wish to avoid a dive into the “trough of disillusionment,” by acknowledging limitations alongside capabilities.

“Let our benchmark be the real-world standards of care whereby doctors grossly misestimate the positive predictive value of screening for rare diagnoses, routinely overestimate patient life expectancy by a factor of 3, and deliver care of widely varied intensity in the last 6 months of life,” the pair urged.

Not a Joke, Per Se
Venture capitalist Chamath Palihapitiya created a stir earlier this year when he blasted IBM Watson on CNBC, calling it “a joke.” “The companies that are advancing machine learning and AI don’t brand it with some nominally specious name after a Sherlock Holmes character,” he said, indicating that it was perhaps the best marketed AI platform, but not necessarily the most advanced.

IBM responded that, “Watson is not a consumer gadget but the AI platform for real business. Watson is in clinical use in the U.S. and 5 other countries. It has been trained on 6 types of cancers with plans to add 8 more this year.” The company closed with, “Does any serious person consider saving lives, enhancing customer service and driving business innovation a joke?”

Ambra Health’s CEO Morris Panner did not quite know what to make of Palihapitiya’s comments, but he did seem to agree on the risk of oversaturating expectations. “It’s not likely that by next Christmas machines are going to be curing cancer. I suspect that Watson sort of made it feel like that reality is already here, when there’s still a lot of work to be done,” he said.

Plenty of healthcare organizations have hitched their wagons to IBM’s supercomputer, hoping that the first and only non-human Jeopardy champion will help find better (and cheaper) routes for diagnosing and treating cancer and other complex diseases.

A 2011 New York Times article published in response to the machine’s game show performance declared that it was “proof that the company has taken a big step toward a world in which intelligent machines will understand and respond to humans, and perhaps inevitably, replace some of them.”

Certainly, that’s the hope or fear for AI, depending on whether one thinks they’re replaceable. Doctors often do not.

Portrait of the Artist as a Supercomputer
A fascinating survey from 2017, conducted by researchers from Oxford and Yale for the Future of Humanity Institute, dove into what machine learning scientists thought the timeline was for the achievement of “high-level machine intelligence.” HLMI for short, they defined the idea as a point in time when unaided machines can accomplish every task better and more cheaply than human workers.

Half of their respondents put the probability of HLMI at 50% half a century into the future. There was a huge split between the researchers surveyed based on their region of origin. North American scientists expect HLMI achievement in 74 years, but Asian scientists foresee it in only 30.

They theorized the ability of AI to perform tasks as well as humans in the future. Perhaps tellingly, the AI researchers don’t believe they themselves can be replaced for 80 years. The consensus placed a computer’s ability to replace the work of a surgeon at about 45 years from now, only a few years further along than AI generating a New York Times best-selling novel.

While the study’s cohort of scientists may see computers soon capable of making actual art, many of the experts who spoke to Healthcare Analytics News think they can’t master “the art of medicine” any time soon.

Medicomp’s Anders, a former internist, is among the doubtful.

“In my mind there is no way that a computer, with a very complex set of algorithms, is going to be able to diagnose and treat illnesses as well as a well-trained, experienced physician. There are a lot of nuances, ‘the art of medicine,’ that the machine is not going to be privy to,” he said.

“I don’t need any assistance in diagnosing and treating the patient in front of me. The flipside is, I don’t believe patients will ever accept [a machine’s diagnosis].”

Clarify’s Drouin, when he puts his patient cap on, reinforces that. “Certainly, it can inform clinical judgment, but if it were my own care I would still want a trained physician to ultimately use the data available to him or her to make a decision based on his or her experience,” he said.

A Glimmer of Precedent
Ambra’s CEO Panner, who is optimistic, would likely agree that humans are irreplaceable, but disagree on AI’s diagnostic and curative potential. He sees great promise that AI systems can go beyond augmentation to emerge as collaborators, all without replacing the work of flesh-and-blood doctors.

“What we’re talking about in the near term is basically process automation and making physicians more accurate and more efficient,” he said. “What is not there today, but could be there, is a computer that for the first time has new ideas.”

The key, in his mind, is getting AI to go beyond merely executing the algorithms fed to them by their makers. He does believe there is a glimmer of precedent, pointing to the example of AlphaGo, the Google DeepMind project that was able to beat an accomplished human player at the ancient board game Go. It did so by outfoxing the human player with moves that defied all conventional wisdom of the game: Essentially, it had developed its own approach.

“It’s obviously a utopian framework, but we’re starting to imagine a world where computers are now true scientific collaborators rather than devices that can augment work mechanically,” Panner explained.

“When somebody comes up with a way to get computers to start finding new ideas to cure disease, that would start executing on Mayo’s vision of ridding the world of illness.”

Until that day comes, physicians’ jobs seem to be secure, and the zenith of modernity’s medical potential is likely still over the horizon.

Become a contributor