Facebook says AI helped reduce hate speech on its platform last quarter

The company said its multilingual systems helped moderate content in several languages including Arabic and Spanish, targeting nearly 27 million piece of hateful content last quarter.

Facebook also said last year it will undertake an independent audit third-party audit of content moderation systems to validate the numbers it publishes.

(Subscribe to our Today's Cache newsletter for a quick snapshot of top 5 tech stories. Click here to subscribe for free.)

Facebook said nearly 97% of the hate speech and harassment content taken down in the final three months of last year were detected by automated systems, before any human flagged it. In the July to September quarter, AI helped detect 94% of hate content; and 80% were spotted in late 2019.

The social network, in its ‘Community Standards Enforcement Report’, noted that in the fourth quarter ending December 2020, hate speech prevalence dropped to about 0.08% of total content from nearly 0.11%.

This means, there were about seven to eight views of hate speech for every 10,000 views of content in Q4, Facebook said in a statement.

The California-based technology company

introduced several artificial intelligence-powered systems last year to help detect misinformation . It started using AI technologies to identify hateful online content in 2016, and has since been adding several updates to its systems which now extends to images and other forms of media.

The company said its multilingual systems helped moderate content in several languages including Arabic and Spanish, targeting nearly 27 million piece of hateful content last quarter.

F acebook has faced criticism previously for its inability to curb hate speech on the platform . Most recently, the social network said it would reduce the distribution of all content and profiles run by Myanmar’s military after it seized power and detained civilian leaders in a coup earlier in February.

Facebook also said last year it will undertake an independent audit third-party audit of content moderation systems to validate the numbers it publishes.