Epidemic as a Dialogue: How AI Learns to Predict Human Behavior ✛
When the pandemic began in France, people stopped meeting each other, and the virus spread more slowly. Scientists wanted to predict how many contacts a person would have if the government introduced new rules. They took a language-understanding program and asked it to read descriptions of restrictions and decide which meetings a person would keep and which they would cancel. These decisions were turned into a table showing how people of different ages interact. Such a table is called an age-specific contact matrix.
The program worked together with the epidemic model: the model calculated how many people got infected and sent these numbers back to the program. The program saw that hospitals were filling up and next time reduced contacts even more. This cycle of mutual influence is called a feedback loop. Previously, scientists had to guess how people would react to restrictions; now the program does it by reading texts. This helped predict hospitalization waves more accurately and assess what would happen if lockdown were delayed by two weeks: contacts would increase by 46% and cases by 45.8%. Such a tool could help doctors and officials quickly evaluate the impact of new measures, even when fresh data on how people interact is not yet available.
During the first lockdown in France, the average number of daily contacts dropped roughly fourfold: from 12.67 to 3.15–3.74. That is like each person, instead of a full bus, talking only to three fellow passengers.
The idea of a model that predicts crowd behavior in response to changing conditions evokes the psychohistory from Isaac Asimov's Foundation, where mathematical models predicted the fates of civilizations. Here, though, the scale is humbler: not a galaxy, but one country and one epidemic.