- Facebook raises its efforts in determining
malicious memes by increasing their AI capabilities
- The social media company created a database of 10,000 memes used to detect and remove “near-identical“ images
- It has also launched the H
ateful Memes Challenge where researchers can win up to USD 100,000
Very soon, “h
ateful memes” will have a stronger detection on Facebook.
Facebook acknowledged that there had been a challenge in determining
maliciously motivated speech due to the limitations of the AI. The tool finds it hard to interpret the combined meaning of texts and images that changes when being put together.
With this, the social media giant announced Tuesday that they would be advancing their AI capabilities to detect na
sty memes. It has created a database of 10,000 multimodal memes; a combination of image and text. They have licensed some Getty images to help researchers with their work.
In a blog by Facebook, it said, “We have a responsibility to keep the people on our platforms safe, and dealing with h
ate speech is one of the most complex and important components of this work.”
As part of its efforts, Facebook launched the H
ateful Memes Challenge for the research community. It is an online competition with a prize pool of USD 100,000.
Facebook’s heavy reliance on AI increased during the coronavirus pandemic when most of its human checkers were suspended.
According to Guy Rosen, Facebook vice president for integrity, said that AI has helped them find more content and they were able to take action on 90% of the material even before they were reported.
He added that a year ago, after the unfortunate event at the mosque of New Zealand, it prompted their commitment to stop organized online inappropriate conduct.
Mike Schroepfer, Facebook’s chief technology officer, said that the system identifies “near-identical” images for it to avoid being reposted.
As shared on Facebook’s blog, AI has detected 88.8% of the h
ate speech content versus 80.2% from the previous quarter. In the first three months of this year, it has removed 9.6 million posts for violating their h ate speech guidelines.