Slashdot Mirror
When Google Got Flu Wrong
ananyo writes "When influenza hit early and hard in the United States this year, it quietly claimed an unacknowledged victim: one of the cutting-edge techniques being used to monitor the outbreak. A comparison with traditional surveillance data showed that Google Flu Trends, which estimates prevalence from flu-related Internet searches, had drastically overestimated peak flu levels. The glitch is no more than a temporary setback for a promising strategy, experts say, and Google is sure to refine its algorithms. But with flu-tracking techniques based on mining of web data and on social media taking off, Nature looks at how these potentially cheaper, faster methods measure up against traditional epidemiological surveillance networks." Crowdsourcing is often useful, but it seems to have limits.
Computer modeling is a powerful technology that should not be underestimated.
However, it should also not be overestimated.
When the "real world" has millions of convergent factors responsible for an event, computer models can sometimes capture a few thousand. Based on those, a simulation is created that suggests a certain outcome. But it may be using less than 1% of the necessary data.
This is like making architectural models out of child's blocks and then being surprised when the building falls down after it is eventually made. There are issues of scale in addition to data that can reveal periodistic or epicyclic patterns that cannot be modeled in a linear method.
They should subtract out a factor based on how much the flu is being talking about in the media.
This is borderline conspiracy think. Scientists of all stripes want their predictions to be testable, with minimal error bars and as accurate as possible.