In a 2008 letter published in Nature, Google researchers claimed the ability, as they titled the letter, to “Detecting influenza epidemics using search engine query data.” Specifically, they claimed to be able to estimate real-time influenza activity throughout the United States. Or, as an article published in Wired seven years later put it, they were “turning the digital refuse of people’s searches into potentially life-saving insights.” Google went on to develop Google Flu Trends, known as GFT.

In a 2008 letter published in Nature, Google researchers claimed the ability, as they titled the letter, to “Detecting influenza epidemics using search engine query data.” Specifically, they claimed to be able to estimate real-time influenza activity throughout the United States. Or, as an article published in Wired seven years later put it, they were “turning the digital refuse of people’s searches into potentially life-saving insights.” Google went on to develop Google Flu Trends, known as

Google had to abandon their project after it missed the peak of the 2013 flu season by 140 percent. By the time the Wired article was written, the rise and fall of GFT had become notorious throughout the tech world as a failed project. The authors of the Wired article warned about GFT as an example of what they called “big data hubris.”

Fortunately, the field of artificial intelligence has greatly advanced since Google’s early attempts to apply its search-engine algorithms to flu epidemiology. A year ago, NBC was able to report “AI is helping turn the tide against flu in two important ways.” What they call “two ways” are actually interrelated aspects of influenza epidemiology: the development of effective vaccines (the second way) relies on accurate forecast of an upcoming flu epidemic (the first way).

As part of its efforts to improve flu surveillance, the U.S. Centers for Disease Control and Prevention (CDC) has continued to evaluate epidemiological forecasting systems which use artificial intelligence. CDC found that among the most accurate are two systems developed by Pittsburgh’s Carnegie Mellon University (CMU). One system, which CMU calls Delphi-Stat, uses AI to make predictions based on past patterns and on input from CDC’s influenza surveillance system. The other, called Delphi-Epicast, creates forecasts from responses from a number of volunteers, a technique sometimes called the “wisdom of the crowds.” During the 2016-17 flu season, CDC scored Delphi-Stat somewhat better on short-term forecasts while Delphi-Epicast was somewhat better on long-term forecasts.

ZDNet reported a study in Finland, Predicting the flu from Instagram, in which an AI analysis of Instagram images found a significant statistical correlation with recorded influenza outbreaks in that country. Investigators gathered weekly Instagram postings from April 2012 to May 2018 based on hashtags in Finnish pertaining to illness, such as the Finnish word “flunssa,” meaning flu, or “lihaskipu,” meaning muscle ache.

Google’s research continues to have relevance. The image processing used in the Finnish study made use of neural networks first developed by researchers at Google in 2016. They trained the networks to look at images of people holding containers such as pill bottles. AI correlated Instagram hashtag in Instagram posts to official incidences of flu recorded by Finland’s National Institute for Health and Welfare.

Finnish study made use of neural networks first developed by researchers at Google in 2016. They trained the networks to look at images of people holding containers such as pill bottles. AI correlated Instagram hashtag in Instagram posts to official incidences of flu recorded by Finland’s National Institute for Health and Welfare. The authors of the Finnish study noted the 2013 failure of Google Flu Trends. Concerned that media attention at the time may have affected their results, they took into account the popularity aspect of social media.

No study of social media would be complete without Twitter. It is hardly surprising, then, that Nature recently reported a study using a Twitter data set and the CDC’s influenza-like illness data set.

Artificial intelligence and social media have come a long way since Google’s failed attempt to promote Google Flu Trends. As established academic institutions continue to develop the capability of artificial intelligence systems, they will become an increasingly important tool in predicting influenza outbreaks.