Facebook has had to quickly train its artificial intelligence to handle an increase in coronavirus content that violates company policies
Facebook said it has removed 2.5 million posts since March 1 offering masks, sanitizers, cleaning wipes and Covid-19 test kits, in an attempt to prevent users from price-gouging or selling counterfeit and dangerous products.
To do it, the social media company relied on the same image-recognition technology it uses to remove sales of guns and drugs. Facebook has had to quickly train its artificial intelligence to handle an increase in coronavirus content that violates company policies, including hundreds of thousands of posts removed for displaying information that could lead to physical harm, like encouraging users to drink bleach as a cure.
The coronavirus statistics released Tuesday are the latest addendum to the company’s twice-yearly report about enforcement to combat the dark side of the way its platforms are used.
With each report, Facebook becomes more reliant on computer solutions to human problems. The company has about 15,000 content moderators working around the world, some of whom can’t work from the office as a protection against the spread of the virus. Between October 2019 and March 2020, about 90% of the content Facebook removed was found first by the software, not user reports, the company said.
In the first quarter of 2020, Facebook said it took down 1.7 billion fake accounts, 39.5 million instances of adult nudity and 7.9 million posts touting illegal drugs.
Facebook said it has been slowly allowing content moderators to do their jobs from home, accounting for privacy and security restrictions, and attempting to monitor their mental health.
Still, long term, the company is investing less in humans and more in AI. Facebook can now build classifiers for harmful content that understand the same concept in multiple languages - which is useful because the company doesn’t have enough moderators proficient in every language its users speak.
“We do unfortunately expect to make more mistakes until we’re able to ramp everything up,” chief executive officer Mark Zuckerberg said on a call with reporters discussing the enforcement statistics.