In 2020, YouTube has more than 2 billion users, with many of them uploading a total of more than 50 million videos.
With more than 500 hours of videos uploaded every single minute, no human can keep up with that growth. This is why YouTube is becoming increasingly dependent on AI-powered automated systems to police its platform.
And in the second quarter of 2020, YouTube said that it took down a record number of videos due to an increased use of its automated systems in its content reviewing process.
In total, 11,401,696 videos were removed from the platform between April and June. And out of that many, 10,849,634 were flagged by its automated systems.
What this means, about 95% of the videos were removed using automation.
AI played an ever bigger role in the removal of user comments. According to YouTube, out of the 2.1 million comments taken down, 99.2% were detected by its automated systems.
According to YouTube‘s Community Guidelines Enforcement Report:
The video-streaming giant company said in a blog post that it needed to rely more than ever on AI due to the impact of 'COVID-19' coronavirus pandemic.
Normally, YouTube uses a combination of human reviewers and machine-learning technology to remove harmful and misleading contents. But since the crisis forces the company to have its staffs work from home, YouTube was faced with two options:
The first option was to scale back its technology, and limit its enforcement to only what could be handled with the company's diminished human reviewers' capacity.
Humans can work more precisely by maintaining a high level of accuracy, as humans can understand things that computers can't. But they are far slower than automation.
This "would result in less content being removed from YouTube, including some content that violates our policies."
The second option was to use YouTube's automated systems on a wider network so that it could catch a lot more content than it previously could.
Automated systems can work non-stop, with the ability to maintain a speed no human can match. The drawback is that, the AI can flag more content than necessary, meaning that YouTube's automated systems can create more false positive.
"With the knowledge that many videos would not receive a human review, and some of the videos that do not violate our policies would be removed," said YouTube.
Since "responsibility is our top priority", YouTube chose the latter, which is more AI and less human.
But acknowledging that its AI that is meant "to help with some of the work normally done by reviewers" can make mistakes, YouTube is also forced to not enforce its rule about the "three strikes and you're out" policy.
Because AI is taking a larger role at the company's content reviewing process, YouTube stopped issuing strikes for content removed without human review - unless it was highly confident that the video broke the rules.
And also because of the possible increased false positives, YouTube has put more staff available to review appeals.
"We’ve always given creators an easy way to appeal if they believe their video was removed in error. Knowing that decisions made by our system are in some cases less accurate than human review, we prepared for more appeals and dedicated extra resources to make sure they were quickly reviewed," said YouTube.
While the number of appeals remain a small fraction of total removals, YouTube saw an increase in the number of appeals and the reinstatement rate to a double if compared to the previous quarter.
YouTube claims that it is only “temporarily relying more on technology” because it simply doesn't have enough manpower to do the job at this moment.
Read: How Coronavirus Forces Internet Companies To Rely More Heavily On AI To Get Jobs Done
