“Any radiologist who plans to practice in 10 years will be killing patients every day,” Silicon Valley investor and Sun Microsystems founder Vinod Khosla told the audience during his keynote talk at the recent Creative Destruction Lab’s Super Session in Toronto.

The long-time AI promoter was making the point he has been espousing for years, namely that due to advances in in artificial intelligence, software will be so effective in the near future that there will be no need for human image interpretation, according to TechCrunch.

“Radiologists are toast,” he told the audience, adding that radiology “shouldn’t be a job,” and that as AI advances continue, humans who continue to diagnose from imaging will “be causing deaths, because [they] choose to practice.”

His position is not new — he has been making the case in public statements since early 2017 — at which time he stated that he thought some types of physicians, including oncologists, could be “obsolete” in five years.

At this most recent address, Khosla said that he now felt the timeline would be more like 15 years.

For long-term AI watchers and medical professionals alike, the timeline for such AI revolutions are not just ambitious, they have a history of repeatedly slipping off into the future.

For some observers, such as computer scientist Geoffrey Hinton, the 5-year prediction was also compelling back in 2016.

“I think if you work as a radiologist, you’re like the coyote that’s already over the edge of the cliff, that hasn’t yet looked down, so doesn’t realize there’s no ground underneath him,” the British-Canadian said in a 2016 video screened at the SIIM-NYMIIS Regional Meeting in New York City in 2017, adding that, “people should stop training radiologists now. It’s just completely obvious that within five years, deep learning is going to do a lot better than radiologists because it’s going be able to get a lot more experience.”

It would seem that despite significant investments and efforts to replace people with machines, calls for the demise of human specialists remain premature.

Back in 2018, for example, the ambitious Watson Health project faced criticism, especially when a Stat article reported on internal company documents said to show “multiple examples” of treatment recommendations that were wrong or unsafe.

In addition, The Wall Street Journal noted at the time that "more than a dozen IBM partners and clients have halted or shrunk Watson’s oncology-related projects."

In response at the time, Dr. John Kelly, IBM senior vice president, Cognitive Solutions and IBM Research, pushed back. He acknowledged that the company had made “a big bet” on healthcare — including Watson for Oncology, Watson for Clinical Trial Matching, and Watson for Genomics — and that although there have been some dropouts, at that time systems were in 230 hospitals and health organizations worldwide.

Progress in AI for health does continue to make progress. At the May American Society for Clinical Oncology 2019 annual meeting, IBM Watson Health unveiled 22 new scientific studies that demonstrated progress in providing clinical decision support for cancer care globally.

"Artificial intelligence technology is helping to enhance the way clinicians treat cancer today, in the real world," said Dr. Nathan Levitan, chief medical officer for Oncology and Genomics at IBM Watson Health in a statement, noting that, "AI is helping multidisciplinary tumor boards make more informed decisions based on curated scientific evidence; it is surfacing critical insights and information that is not identified manually; and it is helping to improve patient satisfaction by delivering a comprehensive view of treatment options."

But that is a far cry from replacing humans in the healthcare work flow, confirming the observations of some experts at the 2017 SIIM-NYMIIS meeting.

“I don’t think radiologists are going to be out of a job or anything close," said Dr. Eliot L. Siegel, a professor and the vice chair at the University of Maryland School of Medicine’s Department of Diagnostic Radiology at that meeting, adding, "I think there’s going to be more radiologists, and I think it’s going to take us about five years to actually start figuring out how do we deliver all this.”

While everyone believes that AI will be part of the future radiology workflow, the more consensus view at present is that AI will shift the burden of big data and raw analysis to machines but will leave human physicians in place to manage the information and do more sophisticated analysis, advised Brady Anderson, the senior director of new product development for enterprise imaging at Philips, at that time.

"There are going to be a lot of algorithms to a lot of this stuff and there are going to be micro solutions that you need to put together," Dr. Keith J. Dreyer, the vice chairman of radiology and director of the Center for Clinical Data Science at Massachusetts General Hospital, said at the 2017 conference, noting, "when those come together, we don’t need to see thousands of numbers but they’re actually accurate numbers. They can go directly to the results of the EHR. But we should also have that access because we might have new data coming to us from this. So, I think people will probably look at this and will add value to the results we put back. This will probably be perceived as radiology, or could be.”