Google Flu

I am at the cutting-edge, part of some kind of avant-garde, an early adopter – all because I got sick. Fever, chills, runny nose – I have the ‘flu. I’m not supposed to though, not yet; at least not according to Google. The information behemoth knows when I (and other US citizens) should be contracting the influenza virus, and they know it just like they know which ads to put on webpages: by aggregating search terms.

Google Flu Trends tracks every time someone searches for the name of a ‘flu medication or types in their symptoms. Based on the individual’s IP address the application can predict where people are getting sick. Launched a few weeks ago, the system works even better than the US government’s own Centers for Disease Control, which depends upon a network of ‘sentinel’ doctors reporting the number of patients with the flu to track the virus’s spread. Only Google can do this faster, cheaper and more accurately. The system was based on five years of search terms and CDC data, which showed that Google Flu Trends accurately predicted outbreaks, besting the federal government by around one to two weeks. A trial version was launched last year and has now been made available to all on google.org – the public corporation’s philanthropic arm. The site is also designed as a tool for public health officials, helping them plan for pandemics and determine where resources should be deployed.

The applications are elegant: a simple blue graph (of course in Google blue) and a clickable map. Designed with the same austerity as Google’s own homepage, the application’s true beauty goes deeper than what can be see on-screen. For example, search terms are dumb, basically small bits of anonymous discrete information that don’t relate to each other. I type in ‘aches and chills’ not thinking that it might impact anyone else; multiplied over millions and millions of searches, however, these terms produce a ‘flu map. It’s a perfect example of that hip trend in computing – ‘swarm intelligence’ or ‘collective intelligence’ – which operates like beehives, flocks of birds and ant colonies. Each animal acts independently, but each independent action taken together creates a ‘super organism’ much greater than the sum of its parts.

In the draft of a paper soon to be published in Nature, the researchers at Google explain, ‘Harnessing the collective intelligence of millions of users, Google web search logs can provide one of the most timely, broad reaching syndromic surveillance systems available today.’ Compare that to market behaviour whereby individual actions actually inflate responses. The first 100-dollar barrel of oil sent a panic through the market as does the news that Asian stocks are trading low. The financial markets work more like an echo chamber in which the individual amplifies a response and that response then amplifies what a person does – a bit like a yawn or cough in a crowded room.

The researchers write, ‘Online web searches, a new form of health-seeking behavior, are submitted by millions of users around the world each day […] 90 million American adults search for medical information online each year.’ Google’s team went through all the search terms in ‘flu season, which included ‘high school basketball’, because the season apparently corresponds with ‘flu season. The search terms were tested with 450 million different models and can be broken down by state whereas the CDC’s data is only broken into eight different regions.

Now if you’re like me, you’re not a little spooked by such corporate prescience or the use to which a search term you believed was anonymous can be put. It raises questions of both privacy and how other organizations – even Google itself – may deploy the data collected in our increasingly well-networked world, in which server logs are repositories of information. Start to combine all that data and we’re not simply talking about losing personal privacy but group privacy. As John Markoff wrote recently in The New York Times, ‘Collective intelligence could make it possible for insurance companies for example to use behavioral data to covertly identify people suffering from a particular disease and deny them insurance coverage.’ One of the experts Markoff sites, Steve Steinberg, a computer scientist working in the financial industry, said, ‘This is one of the most significant technology trends I have seen in years; it also may be one of the most pernicious.’

There is the upside – at least with Google Flu Trends. ‘Flu can cause up to 500,000 deaths a year and in the 1918 ‘flu pandemic somewhere between 20 and 100 million people died. With increased travel and increased poverty due to global warming, the world is primed for another ‘flu crisis. Google wants to create a global version (though I’m unsure of how they deal with areas of the developing world without much computer access), as well as versions specifically dealing with densely populated urban areas.

The research does come with several caveats: for example, Google is unsure how publicity and press attention may affect behaviour and they remain tight-lipped on which search terms they used: ‘While we would like to present the full list of search queries […] upon hearing that Google is using specific queries for influenza surveillance, users may be inclined to submit some of the queries out of curiosity, leading to erroneous future estimates.’ Still, I did my part by Googling ‘muscle aches and fever’, just so they’d know one person in upstate New York was sick. It didn’t change anything: several days later, the incidence of ‘flu is still moderate here.