YouTube is using AI to police copyright—to the tune of $2 billion in payouts

YouTube’s Content ID copyright control system has come a long way since its birth in 2007. After irking some users with its far-reaching, automated approach to addressing copyright infringement, YouTube has since updated Content ID to make it more discerning—thanks in part to Google’s rapidly expanding machine learning technology—and has fine-tuned its controversial payout and dispute mechanisms.  Now, when somebody files a copyright claim against a video, its ad revenue keeps flowing but isn’t distributed to anyone until the dispute is settled. 

Content ID uses audio and visual fingerprinting to detect copyrighted material uploaded to YouTube and lets rights holders decide whether to claim ownership and reap the ad revenue, strip the video of the offending content, or request to have it taken down all together.  At this point, 98% of copyright control on YouTube is handled through Content ID, which has generated $2 billion in revenue for rights holders and creators since 2014. Exactly how that breaks down between the likes of, say, Universal Music Group and up-and-coming artists, YouTube doesn’t say. But suffice it to say that Content ID is providing an automated, scalable alternative to the manual takedown claims that are common under the Digital Millennium Copyright Act. 

In recent months, Content ID has been updated to use smarter fingerprinting that can detect tricks like stretching a video’s aspect ratio, flipping the image horizontally, or slowing down the audio. It’s also been plugged into Google’s machine learning algorithms. In addition to detecting copyrighted video and audio—thanks to a massive database of over 600 years’ worth of reference content provided by networks, record labels, and other rights holders—Content ID can now detect melodies as well.