Facebook’s AI can’t spot Covid-19 disinformation on its own. But with human help, it can slow the spread.

What’s new: Facebook uses a combination of humans and neural nets to crack down on messages that make false claims about Covid-19, which may have deadly consequences. Some of the spurious posts are illustrated above.

How it works: Building a classifier to spot coronavirus disinformation is hard because, as information about the disease evolves, so does the disinformation. Facebook relies on people from 60-plus partner organizations to flag misleading posts and ads, such as false claims that drinking bleach cures the virus, social distancing doesn’t help prevent its spread, and dark-skinned people are especially resistant. Algorithms identify copies of the human-flagged items, even if they’ve been slightly altered.

  • In April, fact-checkers flagged 7,500 false messages about Covid-19. Facebook’s AI systems added warning labels to 50 million related posts.
  • The social network uses a computer vision program to track and label misleading images.
  • Facebook’s data suggests that warning labels deter 95 percent of would-be readers from viewing the flagged content.

Behind the news: The company’s latest models had improved its success rate identifying hateful images and memes. It’s easier to train an algorithm to find hate speech because there’s much more of it than misinformation about Covid-19, Facebook said.

Why it matters: In March, the activist organization Avaaz tracked the proliferation of 100 misleading Covid-19 posts on the social network that it judged harmful for undermining public health measures. The articles collectively were viewed 117 million times. The group said Facebook had taken as long as 22 days to flag some stories. Such delays potentially exacerbate infection rates and lengthen the time before people can gather safely for work or recreation.

We’re thinking: It can be hard even for humans to recognize fakery. But Facebook, as one of the world’s most powerful distributors of information, has a unique responsibility to help its members understand the difference.

Share

Subscribe to The Batch

Stay updated with weekly AI News and Insights delivered to your inbox