A “bug” in Facebook’s News Feed ranking algorithm injected a “surge of misinformation” and other harmful content into users’ News Feeds between last October and March, according to an internal memo The Verge. The unspecified bug, described by employees as a “massive ranking failure,” went unfixed for months and affected “as much as half of all News Feed views.”
The problem affected Facebook’s News Feed algorithm, which is meant to down-rank debunked misinformation as well as other problematic and “borderline” content. But last fall, views on debunked misinformation began rising by “up to 30 percent,” according to the memo, while other content that was supposed to be demoted was not. “During the bug period, Facebook’s systems failed to properly demote nudity, violence, and even Russian state media the social network recently pledged to stop recommending in response to the country’s invasion of Ukraine,” according to the report.
More worrying, is that Facebook engineers apparently realized something was very wrong — The Verge reports the problem was categorized as a “severe” vulnerability in October — but it went unfixed until March 11th because engineers were “unable to find the root cause.”
The incident underscores just how complex, and often opaque, Facebook’s ranking algorithms are even to its own employees. Whistleblower Frances Haugen has that issues like this one are evidence that the company needs to make its algorithms transparent to outside researchers or even move away from engagement-based ranking altogether.
A Facebook spokesperson confirmed to The Verge that the bug had been fixed, saying it “has not had any meaningful, long-term impact on our metrics.”
Still, the fact that it took Facebook so long to come up with a fix, is likely to bolste calls for the company to change its approach to algorithmic ranking. The company recently brought back Instagram’s partially in response to concerns about the impact its recommendations have on younger users. Meta is also facing the possibility of legislation that would like the one used in News Feed.