Facebook said that its AI is now detecting around 95pc of hate speech content on the platform.
As Facebook continues to face scrutiny over how it handles hate speech and potentially harmful content on its social networks, the tech giant has released new content moderation data.
As part of its latest community standards enforcement report, Facebook added a new metric charting the number of hate speech posts on the platform. Between July and September this year (Q3 2020), it estimated that for every 10,000 views of content on the platform, hate speech accounted for between 10 and 11 views, or around 0.1pc.
Over the course of Q3, Facebook said it took action on 22.1m pieces of hate speech content, 95pc of which was spotted by the company before it was reported by a user. By comparison, Facebook took action on 15m posts in Q2.
On Instagram, action was taken on 6.5m hate speech posts during this period, up from 3.2m in Q2. Around 95pc was proactively identified, up from 85pc in Q2, which Facebook said was due to new AI and detection technology for English, Arabic and Spanish languages.
Millions of posts were also removed on Facebook and Instagram for violating policies on violent or graphic content, child sexual abuse, bullying and suicide.
In a blog post, Facebook’s product manager for integrity, Arcadiy Kantor, said: “We’ve taken steps to combat white nationalism and white separatism; introduced new rules on content calling for violence against migrants; banned holocaust denial; and updated our policies to account for certain kinds of implicit hate speech, such as content depicting blackface, or stereotypes about Jewish people controlling the world.
“Our goal is to remove hate speech any time we become aware of it, but we know we still have progress to make. Language continues to evolve, and a word that was not a slur yesterday may become one tomorrow. This means content enforcement is a delicate balance between making sure we don’t miss hate speech while not removing other forms of permissible speech.”
While the number of hate speech posts Facebook took action on increased from 5.5m in Q4 2019 to 22.1m in Q3 2020, the number of removed posts about suicide and self-injury fell from 5.1m to 1.3m during the same period. The number of posts deleted for promoting online hate groups rose from 1.6m to 4m, while the number of violent or graphic posts that were actioned fell from 34.8m to 19.2m.
Data was also released on the number of actioned posts that were successfully appealed or overturned. For example, the number of reinstated posts that were originally deemed as hate speech fell from 665,000 in Q4 2019 to 14,800 in Q3 2020. A significant drop was also seen in the number of reinstated posts previously flagged as child abuse, from 4,400 to 1,300 in the latest quarter.
‘Without our work, Facebook is unusable’
Facebook said that the Covid-19 pandemic continues to disrupt its content review workforce, but it is seeing some enforcement metrics return to pre-pandemic levels. The company said earlier this year that it would be relying more on AI and automation to help moderate its platform during Covid-19 and beyond.
“Our proactive detection rates for violating content are up from Q2 across most policies, due to improvements in AI and expanding our detection technologies to more languages,” Kantor said.
He added that the company is prioritising the most sensitive content for people to review rather than AI, such as suicide, self-injury and child abuse.
However, more than 200 Facebook employees have written an open letter to CEO Mark Zuckerberg and COO Sheryl Sandberg in which they claim they are being asked to return to the office in the middle of a pandemic because the company’s content moderation AI has “failed”.
“Without our work, Facebook is unusable,” they wrote. “Your algorithms cannot spot satire. They cannot sift journalism from disinformation. They cannot respond quickly enough to self-harm or child abuse. We can.”