FACEBOOK SAID IT was making steady progress in curbing hate speech and violent content on the social network as it released its quarterly update on enforcing its standards today.

The social media firm said refinements in artificial intelligence have helped it filter out more hateful and violent content along with bullying, inappropriate sexual material and fake accounts, among other things.

It comes after a Facebook content moderator whistleblower, Isabella Plunkett, recently appeared in front of the Dáil Committee on Trade to discuss Facebook’s outsourced moderation, and the extreme, graphic content moderators have to view.

“My job is to train the algorithm. Facebook’s fantasy is that one-day human content moderators will no longer be required,” said Plunkett.

“That means I’ll get all kinds of content. Hate speech, bullying, graphic violence, suicides, abuse, child exploitation and the list goes on.”

Facebook said the “prevalence” of hate speech in the first quarter of 2021 was between 0.05% and 0.06%,  a drop of nearly half from mid-2020.

Facebook’s vice president for integrity, Guy Rosen, told reporters using this metric of prevalence was appropriate because “it captures not what we took down but what we missed… hate speech prevalence on Facebook continues to decrease.”

The update comes with Facebook under pressure from governments and activist groups to crackdown on hateful and abusive content, along with misinformation and incitements to violence.

The latest update showed violent and graphic content was cut to less than 0.04% on Facebook and under 0.02% on Instagram, and that the company had also reduced the prevalence of other inappropriate content including nudity and sexual material.

The report said Facebook in the quarter removed 25.2 million pieces of “hate speech content,” 8.8 million pieces of “bullying and harassment content,” and 9.8 million including “organized hate content.”

Facebook said it was striving to enforce policies barring harmful misinformation about Covid-19 and had removed more than 18 million pieces of violating content from its platform and Instagram globally since the start of the pandemic.

Refinements to its automated technology have allowed Facebook to make progress in identifying inappropriate content not only in text, but in images and video or even memes which blend these elements, the company said.

“Today we proactively detect about 97% of hate speech content we remove,” Rosen said.

Additional reporting by PA

© AFP 2021