Google Flu Trends Is Wrong. A Lot.
Thursday, March 13, 2014 - 03:40 PM
In 2008, Google launched Google Flu Trends, a service that would track the spread of the flu in the US based on Google searches for symptoms like "cough" or "fever." At the time, journalists heralded it as delivering on the promise of all the data generated on the internet. Well, it turns out that Google Flu Trends is wrong. A lot.
According to a new study, Google Flu Trends has been wrong at predicting the spread of the flu for the last three years, especially during crucial periods of activity. The lead author of the study, David Lazer, says that the tool could likely continue to be effective if Google tweaks the algorithm to change how it weighs certain search terms.
The New Scientist spoke to technology ethicist Evan Selinger, who says the problem is not that the algorithms need to be tweaked, but that they are treated as proprietary by Google, and as such are not open to public scrutiny.
[Selinger] says Google Flu's failures hint at a larger problem with the algorithmic approach taken by technology companies to deliver services we all want to use. The problem is with the assumption that either the data that is gathered about us, or the algorithms used to process it, are neutral. The main concern of tech companies like Google, or data brokers like Acxiom, is to use patterns in that data that can make them money (like where an advert should be placed on screen to maximise the chance that men in Boston aged between 25 and 30 will click it). How they go about that has a huge impact on our lives, and yet we have no idea how it works.
"Algorithmic accountability is one of the biggest problems of our time," Selinger says. "More and more decisions made about us are computed in processes we don't have access to."
Selinger points out that the other areas of life where people make decisions that affect our welfare are much more transparent – we have an opportunity to a fair trial when accused of breaking the law, for instance, and can raise issues with our credit score if we feel it's wrong. In contrast, a business that finds itself pushed to the second page of Google's search results or delisted from Google Maps can never know the reason – the algorithms which make that decision are Google's property.
I know we bang the drum of transparency a lot around here, but if this map is truly meant in service of the public good, Google should at least provide academics a peek under the hood. The better the public understands how this data is gathered and processed, the better we can understand why it provides the results that it does, and how serious those results could be taken. Unfortunately, since algorithms are Google's stock in trade, it is unlikely to be open as I'd like it to be.