Why algorithms need humans to predict the weather
Mari Silbey | September 11, 2012
History is rich with intellectuals who have revered theories of determinism; ideas that suggest if we could only know every facet of a situation, every molecule of the landscape, we could predict and even shape future political, economic, and cultural outcomes.
But when it comes to the weather, forecasters long ago gave up any hope of cataloging all of the variables that could impact rainfall in Seattle, or the arrival of a cold front in New York. At least that’s what Nate Silver reports in his new book, The Signal and the Noise: Why So Many Predictions Fail — but Some Don’t, an excerpt of which was adapted for a recent article in The New York Times Magazine.
If you go by Silver’s account, weather forecasting is something of a dark art. Despite all of the measurements, modeling, and statistical analyses, the weather business relies as much on human insight as it does on computer programming. This is best evidenced by the National Weather Service’s own historical records. According to the agency’s data, a combination of human and computing power creates the most accurate weather forecasts. People improve accuracy levels for precipitation and temperature forecasts by about 25 percent and 10 percent respectively over forecasts done by computers alone.
In other words, the algorithms haven’t bested us yet.
Even as modern futurists envision a time when computers will out-think people, it turns out that there may always be a role for the human mind. In weather forecasting, even the most sophisticated computer modeling systems disagree with each other all the time. It’s up to the people studying those models to illuminate nuance and apply additional context; whether that means knowing how best to weight the variables that determine where a storm is headed, or that morning fog in the northeast tends to dissipate quickly when the wind is blowing in a particular direction.
As powerful as computers are, they can’t “see” everything. And they’re not necessarily as good as humans at knowing when and where to look for more information. Our obsession with big data, and the quantification of industries – finance, advertising, space – can sometimes blind us to the fact that human perception and insight, fuzzy and imprecise though they may be, are still critical to society’s progress. Maybe in the future they’ll laugh at our fixation with numbers. Or maybe they’ll simply recognize better than we seem to that numbers are only part of the equation.
Image credit: Sam Solomon on Flickr
See Also:
Graphic: Business Intelligence Hits the Wall
