Internet

Facebook denies report that its AI fails to detect hate speech

Facebook CEO Mark Zuckerberg   | Photo Credit: AFP

Facebook has denied a report that claimed its AI technology used to detect hate speech or violence has little impact. A narrative that the technology is inadequate and it deliberately misrepresent the progress is false, the company said in a blog post.

(Sign up to our Technology newsletter, Today's Cache, for insights on emerging themes at the intersection of technology, business and policy. Click here to subscribe for free.)

Facebook’s post follows a Wall Street Journal report that said Facebook’s AI cannot consistently identify first-person shooting videos, racist rants and even the difference between cockfighting and car crashes.

The report stated that Facebook employees estimated that the social media giant removes only a single digit percent of the posts that violates its rules.

Besides, according to internal documents reviewed by the Journal, Facebook, two years ago, reduced the time human reviewers focused on hate-speech complaints and made the company more dependent on AI enforcement of its rules.

“Recent reporting suggests that our approach to addressing hate speech is much narrower than it actually is, ignoring the fact that hate speech prevalence has dropped to 0.05%, or 5 views per every 10,000 on Facebook,” Guy Rosen, VP of Integrity said in a blog post.

Also Read | Instagram says social media’s effect on people's mental health “mixed”

Rosen argued that using technology to remove hate speech is only one way to counter hate speech, and if Facebook is not confident that the content meets the bar for removal, the platform may reduce the content’s distribution or won’t recommend Groups, Pages or people that regularly post content that is likely to violate our policies.

He added that when Facebook began reporting metrics on hate speech, only 23.6% of content we removed was detected proactively by our systems and the number is over 97% now.

Rosen noted that Facebook focuses on prevalence since it tells what violating content people see because the company missed it.

“Prevalence is how we measure our work internally, and that’s why we share the same metric externally,” Rosen said.

“While we know our work will never be done in this space, the fact that prevalence has been reduced by almost 50% in the last three quarters shows that taken together, our efforts are having an impact.”


Our code of editorial values

Related Topics
This article is closed for comments.
Please Email the Editor

Printable version | Dec 3, 2021 3:09:35 PM | https://www.thehindu.com/sci-tech/technology/internet/facebook-denies-report-that-its-ai-fails-to-detect-hate-speech/article37052897.ece

Next Story