As the coronavirus pandemic spreads, Facebook's (NASDAQ: FB) automated spam-filtering program is accidentally targeting legitimate posts by news agencies and regular users. The overzealous targeting of legitimate posts has the adverse effect of deleting news posts and informational articles about the coronavirus.
Users on Facebook began noticing that their posts were being flagged as a violation of Facebook's community guidelines on Tuesday, with many users taking to Twitter (NYSE: TWTR) to vent their frustrations. Some users had multiple posts flagged in a row. News outlets such as the BBC and content providers such as Buzzfeed were also victims of the bizarre false-flagging incidents. Particularly concerning was that legitimate news posts and informational posts regarding the coronavirus pandemic were being blocked.
It did not take long for Facebook to notice the bug and take action to correct it. By 6:30 p.m. Facebook announced that the bug had been resolved, though some users were still reporting issues after the announcement. Guy Rosen, Vice President of Integrity at Facebook, responded to the incident, saying "We've restored all the posts that were incorrectly removed, which included posts on all topics - not just those related to COVID-19. This was an issue with an automated system that removes links to abusive websites, but incorrectly removed a lot of other posts too,"
Interestingly, this bug comes just weeks after Facebook revealed the AI behind its automated flagging. Deep Entity Classification, or DEC, sweeps Facebook profiles looking for specific behaviors and "tells" that indicate a fraudulent account. Similar procedures are used to sweep for spam posts on Facebook, with the company employing machine learning so that its auto-moderation AI can learn on its own what to look for in a fraudulent post. Speculation had arisen that the error that lead to the takedowns was the result of machine learning occurring without human oversight, due to many Facebook employees being absent due to the pandemic. Guy Rosen denies that this is the case and only stated that it was a "bug" with auto-moderation, "We're on this-this is a bug in an anti-spam system, unrelated to any changes in our content moderator workforce" he said.