Predicting leaders, storms by computer
By Seth Borenstein, AP
November 13, 2012, 12:25 am TWN
WASHINGTON--Forget political pundits, gut instincts, and psychics. The mightier-than-ever silicon chip seems to reveal the future.
In just two weeks this fall, computer models displayed an impressive prediction prowess.
It started when the first computer model alerted meteorologists to the late October disaster headed for the U.S. Northeast from a bunch of clouds in the Caribbean. Nearly a week later, that weather system became Hurricane Sandy and grew into a superstorm after taking a once-in-a-century sharp turn into New Jersey.
Then, statistician and blogger Nate Silver correctly forecast on his beat-up laptop how all 50 states would vote for president. He even predicted a tie in Florida and projected it eventually would tip to President Barack Obama, which is the equivalent of predicting a coin landing on its side. He did it by taking polling data, weighing it for past accuracy and running 40,000 computer simulations at a time.
He then gave his forecast in terms of percentages, saying that Obama had a 91-percent chance of being re-elected.
In the case of Sandy, lives were at stake. With the election, reputations were on the line and some pundits were dismissive of the computer modeling. Bets were made. Challenges issued.
The math experts came out on top thanks to better and more accessible data and rapidly increasing computer power.
Computers soon should be able to tell health officials where the next food poisoning outbreak will spread, a U.S. government lab predicts.
Tom Mitchell, head of the Machine Learning Department at Carnegie Mellon University, called computer model predictions based on historical evidence "one of the more positive trends we're going to see this century. ... We're just beginning."
Computer modeling tells the government what happens when a nuclear bomb explodes, helped Goodyear make a better tire and helped the makers of Pringles figure out how to keep the potato crisps from breaking in the can, said Bill Tang, program director for the Princeton Plasma Physics Laboratory simulation program.
Every time you swipe a credit card, a computer is using predictive models based on past evidence to determine if it's really you or if it is fraud, Mitchell added.