Facebook is working to set up an AI meme police.
The social media behemoth said it is creating a contest with a $100,000 prize to encourage developers to create artificial intelligence that can “identify multimodal hate speech.”
By multimodal, they mean analyzing the image and words together.
Facebook said it created over 10,000 multimodal examples with images licensed from Getty Images to help researchers. More information on the “Hateful Memes” data set is here.
The competition is hosted by DrivenData
In order for AI to become a more effective tool for detecting hate speech, it must be able to understand content the way people do: holistically. When viewing a meme, for example, we don’t think about the words and photo independently of each other; we understand the combined meaning together. This is extremely challenging for machines, however, because it means they can’t just analyze the text and the image separately. They must combine these different modalities and understand how the meaning changes when they are presented together.
The Hateful Memes data set contains real hate speech. So instead of showing actual memes from the data set, we are instead showing only merely mean examples here. In each of these sample memes, the text phrase and the image are innocuous when considered by themselves. The semantic content of the meme becomes mean only when the text phrase and image are considered together.
The challenge arrives just after Facebook gave plans for a new “content oversight board,” a censorship council that includes anti-Trump activist Pamela Karlan and Muslim Brotherhood supporter Tawakkol Karman, Summit News reported.
Last year, Facebook launched a system called Rosetta to identify inappropriate memes, but Summit News opines it must not have been effective since the company is starting the new Hateful Memes contest.
– – –
Jason M. Reynolds has more than 20 years’ experience as a journalist at outlets of all sizes.