Twitter’s COVID-19 Misinformation Filter Was So Broken It Became a Meme

For a few hours, every single tweet mentioning 5G or oxygen was labeled misinformation.
29 June 2020, 8:47am
Twitter icon

Early Friday morning, Twitter became fun for the first time in months thanks to the introduction of an algorithm that tagged nearly every tweet mentioning “5G” or “oxygen” with a misinformation warning.

Recently, conspiracy theories have been spreading the idea that 5G was causing coronavirus. On the platform, Twitter has been getting users to delete tweets that it believes may cause harm by advocating for unproven treatments or the destruction of 5G cell towers. Sometime before 5 AM until roughly noon, however, Twitter began tagging tweets that mentioned "oxygen" and "5G" with a warning: "Get the facts about COVID-19.”

The misinformation warning would lead to a curated page meant to debunk any possible conspiracy theories connecting 5G to coronavirus and also appeared on tweets that interacted with flagged ones.

The problem was that this warning system appeared to be totally broken. The warning appeared on any tweet that mentioned “5G” and “oxygen” in any combination, even if it really had nothing to do with cellphone towers, conspiracies, or the virus. This led to Twitter users having some fun with the well-meaning but flawed moderation.

None of these tweets are advocating for anything harmful and to a human observer, are obviously a joke. But to a keyword filter or algorithm, each of these tweets presumably threatens to hurt some poor mark reading them. As Kelsey D. Atherton, a defense technology writer, noted in a tweet that ironically got flagged as misinformation: “One of the flaws of attempting moderation at scale by algorithm, a problem that has no bearing on 5G, is that it lets tech companies suck the oxygen out of efforts at reform & regulation, as they shrug & turn ‘we tried one thing with code and it did not work’ into ‘can't be done’”

By noon, the filter no longer seemed to be tagging every single mention of “5G” or “oxygen” even leaving alone tweets that really could be seen as conspiracy theories.

After this shift, Twitter spokesperson Liz Kelly told Motherboard that while Twitter’s algorithm is imperfect and constantly changing based on what’s happening on the platform, it prioritizes over-labeling to err on the side of caution and reduce harm while providing necessary context.

The algorithm may have worked on that front, but the automatic system seems to have exponentially increased the number of tweets about 5G, oxygen, and COVID-19, and now that it’s done has left countless tweets that actually resemble conspiracy theories up without labels. By its own logic, this sort of algorithmic moderation, which is low-cost and hands-off, is deeply flawed.

This article originally appeared on VICE US.