FACEBOOK ROLLS OUT NEW AI TO MODERATE CONTENT: Facebook’s new text-in-image recognition system is live across Facebook and Instagram, according to a company blog post. Codenamed “Rosetta,” the AI system employs tech called optical character recognition (OCR) to scan images for text and then analyzes text to determine the message contained. Rosetta will eventually expand beyond image analysis and flag nefarious text within video content.
Rosetta will facilitate the company’s gargantuan content moderation effort by making it easier to cull through images with text — think: memes — and better detect abusive content, like hate speech. Facebook’s 2.2+ billion users share billions of posts a week in more than a hundred languages, and Instagram just hit 1 billion users worldwide. In addition to AI systems, Facebook contracts about 7,500 people to review over 10 million questionable posts a week. The aim is to review all user-reported content within 24 hours of it being flagged and maintain an error rate below 1%. On a platform of such massive scale, a permissible error rate of 1% still means tens of thousands of moderation errors. In September 2017, Facebook said it would have 20,000 safety and security staff by year-end 2018. But that’s expensive — the more AI can be trusted to oversee content, the more cost efficiently Facebook can handle this issue.