Fb is making AI that may determine offensive memes


Fb’s moderators can’t presumably look by means of each single picture that will get posted on the big platform, so Fb is constructing AI to assist them out. In a weblog submit at the moment, Fb describes a system it’s constructed referred to as Rosetta that makes use of machine studying to determine textual content in pictures and movies after which transcribe it into one thing that’s machine readable. Particularly, Fb is discovering this software useful for transcribing the textual content on memes.

Textual content transcription instruments are nothing new, however Fb faces completely different challenges due to the scale of its platform and the number of the photographs it sees. Rosetta is claimed to be stay now, extracting textual content from 1 billion pictures and video frames per day throughout each Fb and Instagram.

Proper now, it isn’t solely clear what Fb is doing with the information. It’s helpful for fundamental options like picture search and display screen readers, the weblog submit factors out. However it additionally appears like Fb is beginning to put it towards a lot greater objectives, like determining what could be attention-grabbing to place in your Information Feed, and, extra importantly, determining which memes are simply goofy memes and which are literally spreading hate speech or different offensive feedback.

Fb says textual content extraction and machine studying are being put to make use of to “robotically determine content material that violates our hate-speech coverage” and that it’s doing so in a number of languages. Given the corporate’s well-known moderation points, a well-functioning system that may robotically flag doubtlessly problematic pictures could possibly be an actual assist.


log in

reset password

Back to
log in