Facebook AI this autumn began producing forecasts for the COVID-19 pandemic in counties across the nation.
It’s a joint effort with academic researchers and other experts, and one of several initiatives the social media platform has undertaken in response to the pandemic.
The Facebook Artificial Intelligence model reports the cumulative number of positive COVID tests in each county since the start of the pandemic. It’s updated daily and predicts the number for each of the next 13 days.
To judge by the predictions it offered in mid-November for several counties in the greater Capital Region, the model needs some work — it was too low in each county, often by a huge margin. The pandemic has simply accelerated far beyond what the model predicted.
The list below shows the percentage increase in cumulative positive tests projected by Facebook AI from Nov. 17 to Nov. 30 and the percentage increase in actual infections reported by the state Department of Health for that same 14-day period:
- Albany County — 22% — 26%
- Fulton County — 18% — 27%
- Montgomery County — 21% — 46%
- Rensselaer County — 22% — 29%
- Saratoga County — 23% — 34%
- Schenectady County — 20% — 31%
- Schoharie County — 27% — 30%
Facebook AI says that the model uses Facebook’s Data For Good tools, including Symptom Survey and Movement Range Maps.
It says the model uses non-Facebook data as well as public aggregate data to adjust for variables such as mobility, social distancing and symptom prevalence. It attempts to separate regional data from disease-inherent data and account for regional relationships among counties, for example to predict what might happen in one county as a result of an uptick in a nearby county.
The predictions are presented on an online map.
As of Tuesday afternoon, the model is projecting the following increases in cumulative cases through Dec. 13:
- Albany County — 15%
- Fulton County –14%
- Montgomery County — 17%
- Rensselaer County — 16%
- Saratoga County — 16%
- Schenectady County — 16%
- Schoharie County — 15%