POLICING FACEBOOK TAKES time and money, and although Facebook is swimming in the latter, it still has to make do with the same limitations as the rest of us on the former. Not only that, but if Facebook actually paid the sheer number of people required to police the platform - even on cost-saving overseas wages - it would quickly cease to be one of the world's richest companies.
Algorithms to the rescue! Facebook has published a long, detailed blog post on how its been teaching AI to read text on images. This, it says, is not just to aid visually-impaired users of the service, but to spot the difference between a wholesome meme pushing, say, a love of unicorns, and a hate-speech based one pushing a love of white supremacist ideas.
The AI - called Rosetta - is live across the platform now and is claimed to be chomping its way through one billion images and frames of video every day across Facebook and Instagram. This, Facebook says, is just the beginning.
"The rapid growth of videos as a way to share content, the need to support many more languages, and the increasing number of ways in which people share content make text extraction from images and videos an exciting challenge that helps push the frontiers of computer vision research and applications," the company writes.
And that's just the technical challenges - there are also sociological ones. The problem with automatically spotting hate speech on the platform via certain phrases is that people adapt. The language will become more subtle and less overt, and at that point the differences between a wholesome meme and hate speech blur considerably.
Facebook will be hoping that advances in AI mean it can keep pace, but at the very least it would be wise to keep its teams of low-paid Philippino moderation teams doing the most unpleasant job on the planet a little while longer. µ
Looks like someone pressed the wrong button on the routing machine
Half-Life 3 VR anyone
Whilst some old favourites graduate to the main browser