Facebook has revealed that its artificial intelligence (AI) systems are now preemptively flagging 97.1% of the “hate speech” and 49% of the “bullying and harassment” that’s removed from the platform before any human flags it.
In a post discussing its proactive detection of hate speech and bullying and harassment for Q4 2020, Facebook revealed that these numbers reflected an increase from 80.5% of hate speech being proactively flagged by AI in late 2019 and 16% of bullying and harassment being proactively flagged by AI in Q1 2020.
As we’ve reported previously, Facebook has been caught enforcing its hate speech rules selectively and its automated systems are prone to errors. Experts have also explained that there’s no such thing as “hate speech” and that when you attempt to regulate so-called hate speech, the decision-maker can use it as an excuse to ban speech that they hate.
While Facebook often introduces new policies to remove more hate speech and is increasingly relying on its AI to flag posts for hate speech, its transparency report revealed that the amount of so-called hate speech on the platform is tiny.
In Q3 2020, hate speech prevalence was 0.11% meaning that less than one out of every thousand views on the platform contained what Facebook deemed to be hate speech. In Q4 2020, this number dropped to 0.08% meaning that eight out of every thousand views contains so-called hate speech.
While the prevalence has dropped, these statistics show that when looked at in the broader context of everything that gets posted to Facebook, the apparent problem of hate speech that Facebook and the mainstream media often heavily publicize and use as their justification for censoring more content represents a tiny fraction of the total posts.
In total, Facebook removed 26.9 million posts for hate speech, 6.4 million posts for “organized hate,” and 6.3 million posts for bullying and harassment in Q4 2020.
On Instagram, 6.6 million posts were removed for hate speech, 5 million posts were removed for bullying and harassment, and 308,000 posts were removed for organized hate in Q4 2020.
Facebook also vowed to continue improving its technology and enforcement to “to keep harmful content off of our apps.”
The release of these stats follows Instagram announcing yesterday that it’s going to start deleting accounts if they post “hate speech” in their private Direct Message (DM) conversations.