AI's marketing problem: Inflated claims, outsized expectations may actually be slowing AI's move into clinical practice

Share this content:


At Ochsner Medical Center in New Orleans, a team of workers has been outfitted with smartwatches. When a patient is in danger of “coding” (medical argot for a cardiopulmonary arrest happening to a patient in a hospital or clinic), the workers receive an alert on their devices.

Generated by an AI-backed computer model that analyzes patient records, these notifications identify those patients who are at risk of a Code, sometimes hours before the outward signs manifest that a patient is deteriorating so rapidly that their hearts stop beating or they stop breathing.

Early detection allows “a rapid response team to go to these individuals and intervene” before a Code, said Dr. Richard Milani, Ochsner Health System's chief clinical transformation officer. In a 90-day pilot trial, the system reduced adverse events outside of the intensive care unit by 44 percent.

It's just one example of how artificial intelligence has already been integrated into healthcare. Pharma companies such as GlaxoSmithKline are using the technology to speed drug discovery, wearables startups are using it to identify biometric warning signs, and a range of companies, Johnson & Johnson among them, are collecting and mining patient data to build more effective virtual coaching platforms.

These concrete applications are often lacking in discussions of AI's impact on the industry. Instead, the technology is regularly painted in broad, forward-looking strokes. “I have no doubt that sophisticated learning and AI algorithms will find a place in health care over the coming years,” data scientist Andy Schuetz recently told VentureBeat. “I don't know if it's two years or 10 — but it's coming.”

IBM, perhaps more than any other company, trades in big picture promises rather than specifics. According to its website, Watson Health “represents a new partnership between humanity and technology.” With Watson for Oncology, the company said it would tackle one of the most pernicious and complicated diseases: cancer. It was a swing-for-the-fences type of declaration, one that, according to an in-depth report in the medical news site STAT published in September, the company has yet to deliver on.

Watson's underlying technology and workflow have undeniably run into hiccups. But a more central issue is how IBM has chosen to present AI's capabilities, said Leonard D'Avolio, an assistant professor at Harvard Medical School and the founder of a health technology startup called Cyft.

AI is not a “magical oracle that will solve all our problems,” he said, but a powerful tool. “A hammer won't build a house for us, but it's a heck of a lot easier to build a house with one than without.”

When viewed through this lens, better questions begin to emerge, such as: what jobs is AI good at? Its applications are wide and varied, but they typically play out on a much smaller and more specific scale than Watson's goal of revolutionizing cancer care.

No data, no answers.

Using AI to improve operations is a four-step process, D'Avolio says. First, you need access to enough data. Next, you need to structure it. Third, you need to analyze the data for insights, and finally, you need to test, communicate, and incorporate them.

In healthcare, much more than in other industries, the first and last steps are problematic. Data is siloed and often difficult to access. Without enough information to comb through, even the most sophisticated supercomputer can't find relevant insights (according to STAT, Watson has struggled to identify patterns that could lead to more personalized medicine and treatment recommendations for this very reason).

On the other end of the funnel, insights need to be delivered to providers and patients at the right time within the correct contextual framework. In clinical practice, which has many moving parts and literal life-or-death stakes, that's a tall order.

“The visionaries who proclaim that AI is going to transform all of healthcare generally aren't people who are dealing with patients or know about the complexities of managing biosocial illness experiences,” says Brennan Spiegel director of health services research at Cedars-Sinai, who views AI as a tool that will augment, but not replace, most existing healthcare practices.

If steps one and four of the circular process (1, access enough data; 4, test, communicate, and incorporate insights) pose obstacles, it helps to apply AI to use cases where you have a good degree of control of both variables, said D'Avolio. Hospitals, such as Ochsner Medical Center, companies, and startups are already doing exactly that.

All of this isn't to say that cancer care won't be radically improved thanks to applying [AI], or that Watson won't be at the forefront. But the shift won't happen overnight. “We're at the peak of inflated expectations,” Spiegel said. “There is a lot of overpromising, but very few doctors that I talk to think AI is going to change their life right away.”

“Even the grandest vision is achieved in increments,” D'Avolio said. He equates Watson's outsized AI claims to a contractor promising he'll build a house overnight with just a hammer. “Don't blame the hammer, blame the contractor! It's kind of marketing's fault.”


Share this content:
Scroll down to see the next article