Someday soon, your physician may be second-guessed by an artificial intelligence program – and you’ll probably be healthier for it, according to Microsoft Research’s Eric Horvitz.
Horvitz, a research fellow and managing director of Microsoft Research’s lab in Redmond, Wash., laid out the statistics to support second-opinion software during today’s White House workshop on how AI can bring social benefits. The workshop in Washington, D.C., was the second in a series of four sessions aimed at helping the Office of Science and Technology Policy formulate future initiatives on artificial intelligence.
Microsoft Research is pursuing projects in more than 60 areas of computer science, including AI, but Horvitz focused on two projects in particular that brought AI tools to bear on health care challenges.
One project targets medical errors, which Horvitz said are thought to cause more than 400,000 deaths annually in the United States.
“It’s kind of like a city the size of Oakland or Miami going away quietly every year, due to avoidable deaths,” Horvitz said. “It’s the third-leading cause of death in the United States.” (Heart disease and cancer are No. 1 and No. 2.)
Microsoft has been working with partners including the Johns Hopkins Armstrong Institute for Patient Safety and Quality to develop software that scans for potential medical errors. Horvitz said such programs can serve as “safety nets” for health care providers.
“You learn to recognize anomalies,” he said. “You learn to recognize acts of omission and commission and flag them.”
In the future, computers could recognize medical diagnoses that have a higher-than-average probability of taking unexpected twists. “Even experts can be characterized in how their minds work … and build machine-learning models that say, ‘Hey, hang on a second, there’s a significant likelihood you will be surprised by this case’s diagnosis in 48 hours. Take another look,'” Horvitz explained.
AI also plays a role in Readmissions Manager, which started out as part of Microsoft’s Amalga health-care data system and was picked up by Caradigm, a Microsoft-GE joint venture headquartered in Bellevue, Wash. The software is aimed at reducing preventable hospital readmissions, which is estimated to suck up $17 billion in Medicare costs annually.
Congestive heart failure is becoming a major health problem for the over-65 crowd, and it’s posing a problem for readmissions as well. Studies suggest that about half of the Americans hospitalized for heart failure are readmitted within 6 months of discharge. Horvitz said the readmissions typically involve a costly “tune-up” stay that lasts anywhere between eight days and two weeks. But what if early intervention could reduce the need for those tune-ups?
Readmissions Manager sorts through admission patterns as well as medical data to identify high-risk patients and assess intervention strategies. The payoff can be significant, for hospitals as well as for patient health. For example, if you can spend $800 per patient on education programs, follow-up visits or other interventions that reduce readmissions by 35 percent, that will save more than 13 percent in hospital costs.
Horvitz said such efforts deliver on the vision that was sketched out for artificial intelligence back in 1955, when scientists first proposed an effort to “find how to make machines … solve kinds of problems now reserved for humans.” Hhere are some of the other ways in which AI is helping us humans solve our problems:
- Researchers have used software to optimize conservation strategies for a variety of protected species in Washington state’s south Puget Sound region, including the Mazama pocket gopher, the streaked horned lark and the Taylor’s checkerspot butterfly. Horvitz said naturalist E.O. Wilson was impressed by the role that artificial intelligence can play in conservation. “AI may be essential to the survival of life on our planet,” Horvitz quoted Wilson as saying.
- Horvitz and other experts at Microsoft Research and the Seattle Academy of Arts and Sciences harnessed data readings from airplanes in flight to calculate wind speed and direction over a wide swath of the United States. The experiments showed that AI-generated wind modeling can be more accurate than current wind forecasts that are based on measurements beamed down from weather balloons. Better wind modeling, in turn, can lead to shorter airplane flight times and greater fuel efficiency. (You can get the current model from the Windflow website.)
- Microsoft is collaborating with Pivothead on an AI-enabled smartphone/headset interface that lets vision-impaired people know what they’re “looking” at, right down to the text on a restaurant menu. The “Seeing AI” app is still under development, but Horvitz capped off his presentation with a demo of the technology at work:
The White House workshop series began with a May 24 session in Seattle, focusing on the implications of AI for the legal system and public policy. The full video of today’s session in Washington, exploring how AI can contribute to the social good, is archived on Livestream.
Two workshops remain: There’ll be a forum focusing on safety and control for artificial intelligence on June 28 in Pittsburgh, and a session on the near-term social and economic implications of AI technologies on July 7 in New York City.
Insights gained from the workshops will be factored into a White House report to be issued later this year, as well as potential government initiatives on AI.
