Facebook can find nudes but not hate speech

Facebook is still getting to grips with this whole transparency deal. Yesterday, the social network released its first ever public quarterly report on how it enforces community standards, and it appeared that its moderation efforts were fairly effective. But while the social network seemed able to successfully weed out 583 million fake accounts in the first three months of 2018, there was one glaring detail that jumped out in the report.

The company’s auto-flagging AI appears to work well for the detection of fake accounts, spam and even nudity, but it seems to have a lot of trouble detecting hate speech on the platform.

In the report, Facebook said that its AI-based tools were able to take down 98.5% of the fake accounts and nearly 100% of the spam without users reporting the content. The AI even managed to flag 99.5% of terrorist propaganda and 95.8% of adult nudity and sexual activity. It seemed less successful at detecting graphic violence, managing to find only 85.6% of the content, but that figure is massive compared to the amount of hate speech the AI detected. According to the report, the AI was only able to flag 38% of hate speech.

“For hate speech, our technology still doesn’t work that well,” Guy Rosen, VP of product management, admits on the Facebook blog.

“It’s partly that technology like artificial intelligence, while promising, is still years away from being effective for most bad content because context is so important,” he adds. “Artificial intelligence still isn’t good enough yet to determine whether someone is pushing hate or describing that happened to them so they can raise awareness of the issue.”

Language is massively nuanced, so training an AI to recognise something as complex as hate speech, which can present itself in a number of ways, is going to be a tough task without the use of human moderation. Unlike spam and fake accounts, and even nudity and terrorist propaganda – which can be detected through image recognition – hate speech is a lot harder to police.

As Timothy Carone, lecturer at the university of Notre Dame told AP: “It’s not surprising that Facebook’s automated programs have the greatest difficulty trying to figure out differences between permissible opinions and despicable language that crosses the line.

“It’s like trying to figure out the equivalent between screaming ‘Fire!’ in a crowded theater when there is none and the equivalent of saying something that is uncomfortable but qualifies as free speech.”

One solution would be to rely more on humans to gauge whether or not a post can be classed as hate speech, but given the sheer scale of the social network there’s little chance for content to be checked without a level of automation.

In this case at least, automatic detection through AI is not doing enough for it’s 2.9 billion users, and the social network has to find a better way to protect those on its platform. With the wound still fresh after Cambridge Analytica, it may have bigger fish to fry first.