Few U.S. industries are growing as fast as health care, but the big public-cloud companies—Amazon.com, Microsoft, Google—have struggled to crack the $3.2 trillion market. Even as hospitals and insurers collect mountains of health data on individual Americans, most of their spending on extra data storage is for old-school systems on their own premises, according to researcher IDC.
The public-cloud kingpins are trying to lure health-care providers with artificially intelligent cloud services that can act like doctors. The companies are testing, and in some cases marketing, AI software that automates mundane tasks including data entry; consulting work like patient management and referrals; and even the diagnostic elements of highly skilled fields such as pathology.
Amazon Web Services, the dominant cloud provider, is processing and storing genomics data for biotech companies and clinical labs. No. 2 Microsoft’s cloud unit plans to store DNA records, and its Healthcare Next system provides automated data entry and certain cancer treatment recommendations to doctors based on visible symptoms. Google seems to be betting most heavily on health-care analysis as a way to differentiate its third-place cloud offerings. Gregory Moore, vice president for health care, says he’s readying Google Cloud for a world of “diagnostics as a service.” In this world, AI could always be on hand to give doctors better information—or replace them altogether.
The cloud division is refining its genomics data analysis and working to make Google Glass, the augmented-reality headgear that consumers didn’t want, a product more useful to doctors. German cancer specialist Alacris Theranostics GmbH leans on Google infrastructure to pair patients with drug therapies, something Google hopes more companies will do. “Health-care systems are ready,” says Moore, an engineer and former radiologist. “People are seeing the potential of being able to manage data at scale.”
In November, Google researchers showed off an AI system that scanned images of eyes to spot signs of diabetic retinopathy, which causes vision loss among people with high sugar levels. Another group of the company’s researchers in March said they had used similar software to scan lymph nodes. They said they’d identified breast cancer from a set of 400 images with 89 percent accuracy, a better record than most pathologists. Last year the University of Colorado at Denver moved its health research lab’s data to Google’s cloud to support studies on genetics, maternal health, and the effect of legalized marijuana on the number and severity of injuries to young men. Michael Ames, the university’s project director, says he expects eventually to halve the cost of processing some 6 million patient records.
However impressive Google’s AI analysis gets, the health-care industry isn’t exactly a gaggle of early adopters, says James Wang, an analyst at ARK Investment Management LLC. “They can have the lowest error rate and the greatest algorithm, but getting it into a hospital is a whole other problem,” he says. Most electronic medical records are likely to remain “locked inside” health companies for the foreseeable future, says Robert Mittendorff, a biotech investor at Norwest Venture Partners. Indeed, Google’s first major effort in the industry, an online health records service, folded in 2011 because the company couldn’t convince potential customers their data would be safe.
Moore says things have changed since then and that he’s working with Stanford and the Broad Institute, plus about a dozen companies in the health-care industry and defense contractor Northrop Grumman Corp. For now, his primary focus is wrangling more health-care companies onto Google’s cloud, because the more data he can get on Google’s servers, the faster its AI systems will learn. “There literally have to be thousands of algorithms to even come close to replicating what a radiologist can do on a given day,” he says. “It’s not going to be all solved tomorrow.”