There’s no doubt about the Covid-19 emerging on the world stage as a major health emergency. Unlike the past major epidemics, however, technology can provide new tools, both methodological and practical, to contain the virus advance.
Together with standard methods of contagion forecasting, scientists and startups are working on new algorithms that combine and integrate local mobility data, international trade and even the analysis of keywords in social networks: Big Data for public health. With the advent of social networking platforms, we now get to know in-depth the contact networks that shape social interactions and simulate computer epidemics.
The main challenge, however, is to integrate heterogeneous information sources before computers can even process them.
One example is the HealthMap system, designed by Boston Children's Hospital, which periodically scans news and social media, producing constantly updated and geolocalized reports on the global health situation through its AI engine. On December 30, 2019, the system had noticed an abnormal viral outbreak in the city of Wuhan, assigning a risk level three out of five, which went unnoticed by the experts. Or, on the diagnostic front, Alibaba funded an developed a tool that can diagnose coronavirus pneumonia by analyzing CT scans in just over twenty seconds and providing a 96% accurate report.
A reliable early warning system is what governments around the world now need most to manage emergencies. Several startups now provide such tools, integrating information from a variety of computer sources, from public transport flows to cell phone connections.
However, entrusting public health choices to automated systems remains an open point of discussion within the scientific community. Models, however accurate they may be, need quality data, which require a continuous process of comparison, often supervised by expert groups.
Finally, the close interaction between models and real experience also serves to avoid unintentional systematic deviations towards biased predictions, a problem known as model bias. A model affected by inaccurate or systematically distorted data can produce overly polarized reports, with the risk of underestimating the severity or spreading panic around.
Artificial intelligence can largely extend human capabilities, but it remains our task and responsibility to use it for the benefit of all.