The Facebook logo. - AFP

While the pandemic has limited its human capacity to moderate content on its platforms, Facebook is increasingly using artificial intelligence (AI), including to detect "memes" carrying hate messages. If the algorithm only understands words, messages are more difficult to be perceived.

"The AI ​​now proactively detects 89% of the hate content we remove, compared to 80% in the previous quarter," the American group noted on Tuesday. "In the first quarter of 2020 alone, we acted on 9.6 million content that violated our rules in this area, or 3.9 million more."

The task is harder on Instagram

As a sign that the task is more difficult on Instagram, this proactive detection rate drops to 44% on the application of photos and videos, according to Guy Rosen, vice-president of the Facebook group. The teams of the global social network have made a database of 10,000 memes available to researchers, so that they can develop algorithms capable of locating problematic messages.

Facebook has also reported on its own progress: its AI tool SimSearchNet is said to be able to recognize copies of an original image, despite possible modifications to divert it from its original meaning and evade automatic detection. Linguistically, the platform's algorithms can now "understand" the same messages in different languages.

Counterattacks, which sometimes use the words of the insulting message, "are particularly difficult to identify correctly because they can seem very similar to hate speech." Heather Woods, a professor at Kansas State University and specialist in memes and extremist content, welcomed Facebook's initiative to include outside researchers to tackle this complex problem. "The nuances of memes and contextual specifics will remain a challenge for platforms looking to get rid of hate content," she comments.

High-Tech

Messenger Rooms, the competitor of Zoom launched by Facebook, will land on WhatsApp

High-Tech

Facebook presents the Council of Elders who will be the ultimate arbiter of moderation

  • High-Tech
  • Bullying
  • Facebook
  • Artificial intelligence
  • Same