Stanford researchers: Bring expectations for artificial intelligence back down to earth

AI
Healthcare needs to recalibrate its expectations of AI and find ways for the technology to work with physicians.

Artificial intelligence is hitting its stride—at least when it comes to hype.

The unbridled excitement surrounding AI and machine learning technology is higher than ever before, so much so that it’s become a distraction for the medical community, according to two Stanford researchers.

Arguing that AI has reached the “peak of inflated expectations,” Stanford researchers say the healthcare industry needs to shift its focus to how rapidly evolving technology can improve care.

Arguing that AI has reached it’s “peak of inflated expectations,” Jonathan H. Chen, M.D., and Steven M. Asch, M.D., from Stanford University’s Department of Medicine wrote in the New England Journal of Medicine that healthcare can "soften a subsequent crash into a 'trough of disillusionment' by fostering a stronger appreciation of the technology’s capabilities and limitations."

One particularly radical example emerged earlier this month when venture capitalist Vinod Khosla said AI would soon render oncologists obsolete.

“I can’t imagine why a human oncologist would add value, given the amount of data in oncology,” he said during an event in San Francisco hosted by MIT, according to VentureBeat. “They can’t possibly comprehend all of the things that are possible.”

RELATED: AI’s biggest challenge will be getting along with physicians

That's the kind of debate that is viscerally satisfying, but ultimately ineffective, the authors contend. Instead, healthcare would benefit from an approach that combines the talents of both humans and algorithms. Healthcare informatics experts took to Twitter on Wednesday night in support of that approach.

AI’s true value will be in decision support, Chen and Asch added. Current AI applications often reach conclusions that are already well known to patients. For AI to make an impact on clinical care, the technology needs to provide predictions that can influence care decisions enough to improve current practice.

“Before we hold computerized systems (or humans) up against an idealized and unrealizable standard of perfection, let our benchmark be the real-world standards of care whereby doctors grossly misestimate the positive predictive value of screening tests for rare diagnoses, routinely overestimate patient life expectancy by a factor of three and deliver care of widely varied intensity in the last six months of life,” the researchers wrote.

But first, the industry will have to tackle a significant hurdle in making patient data more accessible and available in real time. As Chen and Asch point out, clinical data for predictive purposes has a shelf life of about four months.

Healthcare systems are undoubtedly invested in AI, although the technology’s practical implementation is still fairly limited. Partners HealthCare recently launched a 10-year partnership with General Electric to develop and commercialize AI platforms for the industry. Academic medical centers like the University of California San Francisco and the University of Chicago are teaming up with Google to build on initiatives involving predictive analytics.