Facebook bots flagging offensive images faster than users

1 Jun 2016

Facebook has claimed its AI is already far better at flagging offensive content than its users are, just days after it agreed to new EU hate speech regulations.

Yesterday, we reported that Facebook, Twitter, Microsoft and YouTube had agreed to new European regulations that will require them to combat illegal online hate speech on their respective platforms.

Now the former has claimed its AI is already pretty good at spotting posts that should be removed, better than Facebook users, in fact.

Speaking to TechCrunch, Facebook’s Joaquin Candela said he found it “interesting” that the company’s AI was flagging offensive photos at a faster rate than humans.

“The higher we push that to 100pc, the fewer offensive photos have actually been seen by a human.”

Considering the fortunes that tech companies pump into AI and machine learning with a view to better advertising, it makes sense that this can be flipped around to produce results in other areas.

Under the EU agreement mentioned yesterday, companies essentially signed up to a new code of conduct that will see them take the lead on stopping the spread of illegal hate speech online.

This involves putting in place processes to review notifications regarding illegal hate speech so they can swiftly remove or disable access to such content within 24 hours.

They will also be obliged to educate and raise awareness among users about the type of content not permitted on their platforms.

If offensive images are being caught before most users see them, though, the future codes of conduct in this area may relate to less and less content.

Of course, offence is not always universal, it can be very subjective, as Facebook found out to its cost recently when it removed an advert for going against its policies.

It blocked an ad that showed plus-size model Tess Holliday in a bikini that, it said, “depicts a body or body parts in an undesirable manner”.

In a wonderful irony, the ad was promoting positive body image, with the backlash swift and powerful enough to force Facebook to reverse its decision.

Facebook image via Silatip/Shutterstock

Gordon Hunt was a journalist with Silicon Republic

editorial@siliconrepublic.com