“Here’s what Meta is actually saying: Sure, they might identify and remove only a tiny fraction of dangerous and hateful speech on Facebook, but of that tiny fraction, their AI classifiers catch about 95 – 98% before users report it. That’s literally the whole game, here.”
Meta’s internal memos put hate speech removal at 3 to 5% and violence and incitement removal at about 0.6%. Its public reports brag that 95% of removals are “proactive.” Do the math and the classifiers catch about 4.75% of hate speech. Reporters at Fast Company, Axios and Bloomberg repeated the 95% figure anyway. Meta made $40 billion in profit in 2021, and the stat it sold the press is a percentage of its own failure.