Turns out Google’s not so hot at predicting flu cases
Google is great at answering queries about when the Super Bowl starts and why blueberries are so good for you, but it has yet to figure out how to accurately predict cases of the flu. It seems that Google Flu Trends, which launched to much hype in 2008, has been off the mark, says Northeastern University’s David Lazer — the lead author of a new study in Science that analyzed Google’s predictions.
“It missed by a huge amount last year and actually, it turns out, it’s been missing by a fair amount for several years,” Lazer told NPR.
Google flu trends works by identifying certain search terms that would be good indicators of flu activity. It then estimates how much flu is circulating in a given area. But Lazer’s team found that for the 2012-2013 flu season, Google predicted twice as many doctors’ visits for flu as the Centers for Disease Control and Prevention eventually recorded. In 2011-2012 it overestimated its predictions by more than 50 percent.
Why does this matter? Transparency for one reason, says Evan Selinger, a technology ethicist at Rochester Institute of Technology in New York. When a large company like Google uses data it gathers about us, and shares that information in a way that will impact our lives, we should have access to how that system works. But Google does not disclose information about its algorithm.
“Algorithmic accountability is one of the biggest problems of our time,” Selinger told New Scientist. “More and more decisions made about us are computed in processes we don’t have access to.”
In 2013, Google released a report that addressed certain elements of its algorithm for tracking flu and dengue fever, stating that predictions had been susceptible to heightened media coverage. Google updated its formula last fall to reduce some of those instances.