The coronavirus pandemic already had a significant impact on how companies work across the world. Google’s YouTube is putting its expertise in the AI to good use in the crisis. According to the company’s latest Community Guidelines Enforcement Report, YouTube has removed the most videos in a single quarter during Q2 2020.
Out of the 11.4 million videos removed from its platform, AI moderators have flagged 10.85 million videos. Google is letting its employees work from home until July 2021. With the reduced workforce, YouTube has chosen to use its automated systems for removing the harmful content from its platform.
YouTube automated systems removed record videos in the last quarter
With automated flagging, the company can also act fast. As per the report, 42-percent of the video removals happened even before getting any views. Obviously, this also resulted in some videos getting removed even without violating the company’s policies. However, the option of the human review will be a better choice in similar scenarios.
It also helps in training the machine learning systems for increased accuracy in the future. Nevertheless, automated systems usage helped the company remove double the number of videos than the previous quarter.
This is also the first full quarter to use only AI-based moderators
A majority — 33.5-percent — of the videos removed included child safety issues. While another 28.3-percent were flagged as misleading or spam, there are 14.6-percent videos with nudity or sexual content. Speaking of the user comments, the machine learning systems removed 99.2-percent of the 2.1 million comments during the same period.
This is also the first full quarter the company ever operated with all algorithm-based content moderators. Regarding region-wise video removals, the US and India topped the list with 2,061,733 and 1,446,772 videos respectively.
Since the company started relying on automation, there is also a spike in the claims from the creators. Compared to its previous quarter, the claims also doubled up in the quarter ending June 30.
The company is also accepting a lower level of accuracy for videos covering certain sensitive policy areas like violent extremism and child safety. While this helps remove more videos as quickly as possible, many of them might not even violate the company’s policies. During the same period, the violent extremism content on YouTube also saw four times the growth.
The success of machine learning will mainly depend on the accuracy it brings with it. Till then, more and more video removals will happen just because of the algorithm errors.