Have you ever had to moderate a conversation or content? You might have a comments section on your website and if you have a social media account people have the ability to comment on your work. You may have had to moderate a conversation on that or you may have pulled up a story and read the comments. It can get nasty. Have you ever wondered just how much of a job moderating Facebook (and Instagram) is?
Facebook recently released its Community Standards Enforcement Report for 2019 through the month of September. The results are staggering.
What Is Included?
Among the content that is heavily moderated are of course many things that you would expect like sexually explicit material, drugs, terrorism, violent content and harassment or bullying. Also included are more nebulous things that may have been more controversial like spam and hate speech. They also detailed how many fake accounts (or at least known fake accounts) were removed during this time.
How Much Was Removed?
Adult content like nudity saw over 52 million pieces of content removed with nearly all of it being detected by Facebook’s AI system. 10.3 million pieces were appealed and around 1.8 million pieces were restored. Around 24 million pieces of child nudity or child exploitation were removed with again almost all of it being detected by AI. Around 300,000 pieces were appealed and around 35,000 pieces restored.
When it comes to bullying over 8 million pieces of content had action taken on it and in this case most of it was flagged by users and not detected by AI. About 1.5 million pieces were appealed and 225,000 pieces restored. Bullying can lead to other things like suicide or self harassment and Facebook has stepped up its enforcement in this regard. Data included only stretches back to April but in that time around 4.5 million pieces had action taken on it and almost all of it was detected by AI. Around 90,000 items were appealed and about 12,000 items were restored.
Then there is of course the more controversial cause of hate speech. Around 8.5 million items were actioned with a little over 50% detected by AI. In this case about 3.6 million items were appealed and a little over 400,000 were restored.
There is more of course. The link to the report is above or you can click this here as well to go to it.
How Bad Are Fake Accounts And Spam?
No one of course likes spam and of course Facebook is not alone when it comes to the war on spam. The report says that Facebook blocked over 4 billion pieces of spam. Around 48 million of those pieces were appealed and 250 million pieces restored. In this case most of it was restored without an appeal. Is Facebook’s spam detector overzealous? Maybe. Just proof AI may not be perfect after all.
Now how many fake accounts were removed? 3.7 billion of them. Considering the world population is believed to be around 7.7 billion that is a lot of accounts and how much this cuts into Facebook’s claimed 2.45 billion users worldwide has yet to be seen. Either way that is a lot of fake accounts but considering how easy it is to create an account and harvest the information of people it’s not surprising. You’ve probably seen a fake account at some point and hopefully you did not accept their friend request.
Chances are as well that human moderators would not be able to keep up with this number of fake accounts. AI is necessary to detect them and remove them. One can imagine how long it might take to examine an account, determine if it is fake and then remove it with a human doing the work. In this case AI is certainly more efficient.
But What About Instagram?
Facebook published for the first time data about their efforts on Instagram in this report. Instagram had nowhere near the numbers as Facebook but over 1 million pieces of content of adult nudity, 3 million pieces regarding regulated good like drugs or firearms, 1.8 million pieces related to suicide or self-injury, about 240,000 pieces of terroristic content were actioned.
Considering Instagram has an estimated 111 million users that is probably not as bad as expected. It may just be a matter of time and of course since Facebook is reporting these issues for the first time there is nothing to compare these numbers to.
Reflections
Of course this is what Facebook knows about, either by being detected by their AI system or by user reports. It is probably safe to say that the system was not able to detect everything and savvy users like terrorists (over 12.5 million pieces deemed to be terroristic content were actioned in 2019 so far) may have numerous workarounds.
At the same time one of the major complaints about users is regarding hate speech. What is considered to be hate speech to one person may not be another and Facebook’s terms of service does little to clarify things. Given the amount of content that was restored it just shows that AI has some work to do and that not every piece flagged is truly hate speech. It is admittedly an uphill battle that Facebook cannot win easily, if at all.
The good news for many Facebook employees is that AI is handling much of the content moderation, which was not always the case. Being a Facebook moderator is one of the most mentally damaging jobs out there and they review some of the worst that humanity has to offer. Of course AI is not perfect but it hopefully has released some of the burden on those moderators.
Conclusion
For most of this content no one will have an issue with it being banned. Very few people will be OK seeing sexual acts pop onto their Facebook feed and no one should want to see children exploited in that manner. Facebook has of course been coming under fire from all sides regarding their policies and their actions. Maybe they are doing something right if they have managed to anger both sides. Or maybe not.
Everyone should recognize that the job that a social media platform that is open to just about anyone and everyone on the planet will require moderation. Of course there are different cultures on this planet and even in this country. True across the board moderation is extremely difficult and the numbers published in this report is just the tip of the iceberg.