The first international “bell” on the Covid-19 was rang not by a human but by an American computer with artificial intelligence.
On December 30, 2019, the “smart” software of the Boston Children’s Hospital HealthMap website, which uses artificial intelligence to monitor social media, news, online searches and other information that may indicate signs of a new epidemic, was the one that first detected a report of a new type of pneumonia in Wuhan, China. The program immediately notified the recipients of a list via e-mail that seven people were in critical condition and assessed the severity of the situation in number “three” on a scale of five.
But people weren’t late anyway. Doctors in Taiwan had already informed American epidemiologist Marjorie Pollack in New York that the issue has begun to be discussed on the popular Chinese social media Weibo. Pollak, according to Science, recalled that something similar had happened in 2003 at the beginning of the outbreak of the severe acute respiratory syndrome (SARS).
Nearly an hour after HealthMap’s automatic warning signal, Pollack made a more detailed post to the Emerging Diseases Surveillance Program, which informs about 85,000 people internationally. And so the rest of the world began to turn their attention to something strange that was happening in China.
The “bell” from HealthMap shows the potential of artificial intelligence to monitor epidemics. As the Covid-19 pandemic continues to spread internationally, artificial intelligence researchers are trying to “set up” automatic detection and warning systems, which will search for fleas in haystacks, amid huge volumes of online data, to discover the first signs of a new epidemic outbreak in some part of the world.
Artificial intelligence is not going to replace traditional human epidemiological surveillance, at least not yet.
“We should regard it as another tool in our quiver. I don’t think it can replace the test.”
said epidemiologist Matthew Biggerstaff of the U.S. Centers for Disease Control and Prevention (CDC).
Already, well before Covid-19, the U.S. CDC had since 2013 launched an annual competition for the most accurate system for predicting the severity and spread of a new flu wave in the U.S. Dozens of suggestions are filed every year, and about half involve machine learning algorithms (artificial intelligence), which recreate google searches, Twitter tweets, Facebook posts, Wikipedia websites, etc., to find the earliest signs of an impending epidemic or outbreak.
Many of these research groups, which have hitherto been involved in influenza, have now turned the tools of artificial intelligence in the direction of Covid-19. Their aim is both to assess the current state (now-casting) of the epidemic and to predict its course (forecasting).
It’s not an easy job, nor has it guaranteed success. Between 2009-2015 Google used a new “smart” tool, Google Flu Trends (which is now integrated into HealthMap), to search users’ searches and to detect the first -still undisclosed to scientists- signs of a new flu wave. At first the system did well and predicted the evolution of the flu two weeks before the CDC.
But then it overestimated the spread of the flu, “unnecessarily terrorizing” epidemiologists. The main cause of the failure was that Google researchers did not “retrain” their system to take into account changes in user behavior during their searches, thus misinterpreting, for example, searches for flu-related news as evidence of infection in the user itself.
So skeptics believe that artificial intelligence, which is constantly improving, will be better prepared and useful in the next pandemic. What is certain, however, is that artificial intelligence came to be an instrument of epidemiology.