IE 11 Not Supported

For optimal browsing, we recommend Chrome, Firefox or Safari browsers.

Did Facebook help build an AI to predict COVID outcomes?

Answer: Yes.

A rendering of a digital brain on a white background.
Shutterstock/Sergey Nivens
Facebook’s AI division, known as FAIR, and NYU Langone Health have built a set of machine learning algorithms that can predict how a patient who has been infected will respond to the coronavirus.

It works by examining chest x-rays, either a single image or a sequence, and then predicting the level of patient deterioration up to four days in advance. It can also determine how much oxygen the patient will likely need.

The team, which began its work on the AI back in the spring, started training it initially on publicly available chest x-ray databases. Since the virus was so new, however, there were virtually no x-rays of patients who had it — certainly not enough to create a sufficient data set for training an AI. Therefore, the team employed a method called “pretraining” in which they trained the AI initially on non-COVID chest x-rays, and then fine-tuned it at the end with the small number of COVID-19 chest x-rays that had been made available.

The tool, therefore, doesn’t work as a method of diagnosis — it can only tell us how the lungs of a patient with COVID-19 will be affected by it. “We're not making the diagnosis of COVID — if you have COVID or not — based on an x-ray,” said Facebook AI program manager Nafissa Yakubova. “We are trying to predict the progression of how severe it might be.”