Facebook uses AI to forecast COVID-19 spread across the US

It could help counties take measures in time to minimize losses.

Beata Zawrzel/NurPhoto via Getty Images

It’s crucial to predict the spread of COVID-19 during the pandemic when a poor estimate could lead to overcrowded hospitals or extra-strict lockdowns. Facebook is betting that AI could help. It just published a paper outlining an AI technique it will use to forecast the spread of COVID-19 in counties across the whole US. The system predicts infections 14 days ahead by accounting for both the nature of the disease and the social factors that influence its reach.

Facebook factored in the new coronavirus’ inherent traits, but also trained its AI on time-based county case data as well as public, anonymized data that helped it account for elements like mobility and social distancing. The company also crafted a “neural autoregressive model” meant to separate regional and disease-specific elements of those data sets. A spike in one county could affect forecasts for nearby counties, for example.

The researchers believe their system could be more effective not just by offering a more complex model, but by making relatively few assumptions.

The social media giant is publishing updated forecasts every week on the Humanitarian Data Exchange. It also hopes to translate its methods to Europe through a team-up with Barcelona’s Universitat Politècnica de Catalunya. The US forecasts will be included in European Commission-oriented reports to help understand global COVID-19 hotspots.

Whether or not this will prove useful stateside depends on adoption, not to mention real-world accuracy. If it works as promised, though, it could help officials adjust hospital capacity and implement milder restrictions. Vaccines and reliable treatments are the ultimate ways out of the pandemic, but Facebook’s AI could keep conditions more bearable until that moment.