Facebook on Tuesday unveiled an initiative to tackle "hate memes" using artificial intelligence, backed by the public offering, to identify maliciously motivated posts.
The main social network said it had already created a database of 10,000 memes, images often mixed with text to deliver a specific message, as part of an accelerated effort against the speech of hatred.
Facebook said it was launching the database to researchers as part of a "Hateful Memes Challenge" to develop improved algorithms to detect hateful visual messages, with a price of $ 100,000 (around Rs .75.4 lakh).
"These efforts will stimulate the wider AI research community to try new methods, compare their work and compare their results to speed up work to detect multimodal hate speech," Facebook said in a statement. blog post
Related Downloads:
Facebook's effort comes when it relies more on artificial intelligence to filter objectionable content during the coronavirus pandemic that has sidelined most of its human moderators.
Its quarterly transparency report indicates that Facebook deleted approximately 9.6 million messages for violation of "hate speech" policies in the first three months of this year, including 4.7 million "hate-related" content. ".
Facebook said AI was better tuned on filtering as the social network turned more to machines after the crashes.
Guy Ro sen, vice president of integrity for Facebook, said that with AI, "we can find more content and now we can detect almost 90% of the content we delete before someone tells us. "
Facebook has said it has pledged to "disrupt" hate behavior organized a year ago after the deadly attacks on mosques in New Zealand that prompted governments to "call for" action "to curb the spread of online extremism. .
Automated systems and artificial intelligence can be useful, Facebook said, to detect extremist content in multiple languages and analyze the text embedded in images and videos to understand their full context.
Mike Schroepfer, Facebook's chief technology officer, told reporters during a conference call that one of the techniques that helped this effort was a system to identify "almost identical" images, to deal with posting malicious images and videos with minor changes. to escape detection.
"This technology can detect near-perfect matches," said Schroepfer.
Heather Woods, a professor at Kansas State University who studies memes and extremist content, praised Facebook's initiative and the inclusion of outside researchers.
"Memes are notoriously complex, not only because they are multimodal, incorporating images and text, as Facebook points out, but because they are contextual," said Woods.
"I imagine the nuances and contextual specificity of memes will continue to be a challenge for Facebook and other platforms seeking to eliminate hate speech."
Is my 10 an expensive OnePlus 8 or a budget S20 Ultra? We talk about it on Orbital, our weekly technology podcast, which you can subscribe to via Apple podcasts