Intellectual Property

Takedowns Over Classical Composers Highlight Problems Of Automated Content Filtering

Content ID systems don’t understand limitations and exceptions, like fair use, parody, or criticism, and will flag content even if it’s not infringing.

By  
on

I briefly mentioned the problems of automatic filters and content ID systems in a post discussing the new EU Copyright Directive, but let’s take a deeper dive into some real world examples.

Here’s a brief explanation of how content ID systems works. The hosting platform creates a database, which rightholders contribute to, of audio and/or video. The content ID system checks against content that is uploaded to the platform to see if there is a “fingerprint” match that exists in the database. Under YouTube’s content ID system, when a video is uploaded, the system flags videos that match the “fingerprint.” Let’s say you upload a video of your toddler dancing and about 15 seconds of a well-known song is playing in the background — it doesn’t matter what you title the video or how you describe it or even what the video shows — the ID system will catch the “fingerprint” of the song (see the “Dancing Baby” case). YouTube then allows the purported rightholder to decide whether to block the video (i.e., have it taken down), track it, or monetize it.

While this system may seem reasonable — and I understand that in the digital world, rightholders are rightly concerned that infringing content will be posted and spread quickly across the Internet — automatic filters have several problems. First, rightholders can contribute to the content ID databases with material that’s in the public domain or not protected by copyright. The ID system probably won’t be able to tell that it’s in the public domain, so anything matched against it will be flagged. Additionally, content ID systems don’t understand limitations and exceptions, like fair use, parody, or criticism, and will flag content even if it’s not infringing. Unless these automated matches are reviewed by actual humans (and even then, there’s no guarantee that the right result will occur), content ID systems won’t be able to adequately address limitations and exceptions. Large platforms have an incentive to err in the side of over-inclusiveness in content ID matches and will capture non-infringing uses.

To illustrate an example of content ID system failure, a German music professor had public domain recordings of Beethoven’s 5th Symphony taken down. Professor Ulrich Kaiser digitized records, taking care to only digitize recordings where the copyright term had expired. In one video, he explains his project — intended as part of an effort to promote Open Educational Resources (OER) — with sample clips of the music played in the background. After uploading the video to YouTube, within three minutes, he received a Content ID notification, but was able to successfully challenge the claim. Intrigued, he decided to run a test, creating a new YouTube account (“Labeltest”) to see what other non-copyrighted music might be flagged. As he notes on the Wikimedia blog:

I quickly received ContentID notifications for copyright-free music by Bartok, Schubert, Puccini and Wagner. Again and again, YouTube told me that I was violating the copyright of these long-dead composers, despite all of my uploads existing in the public domain. I appealed each of these decisions, explaining that 1) the composers of these works had been dead for more than 70 years, 2) the recordings were first published before 1963, and 3) these takedown request did not provide justification in their property rights under the German Copyright Act.

I only received more notices, this time about a recording of Beethoven’s Symphony No. 5, which was accompanied by the message: “Copyrighted content was found in your video. The claimant allows its content to be used in your YouTube video. However, advertisements may be displayed.”

Kaiser’s experiment wasn’t the first time Content ID systems have had a run in with classical composers. James Rhodes uploaded a video of himself playing a song by Bach on Facebook and was hit with a takedown notice by Sony, which claimed ownership of 47 seconds of the song. Here we’re not even talking about a recording whose rightholders might accidentally be infringed upon, because it was Rhodes’ own video of himself. When Rhodes filed a counter-notice to get his video restored (in which, he apparently explained, “This is my own performance of Bach. Who died 300 years ago. I own all the rights.”), Sony rejected the counter-claim until Rhodes took to Twitter.

While we’re on the topic of artists’ own videos being taken down or flagged by content ID systems, musician Gavin “Miracle of Sound” Dunne’s YouTube channel was hit with seven copyright claims for his own songs. Who filed these claims? Dunne’s label did, apparently on Dunne’s behalf.

Sponsored

And what about U.S. government works that are in the public domain? They’re not exempt from being challenged, even on their own YouTube channels. NASA’s video of the Mars landing of the Curiosity Rover was flagged on its own channel, despite that 1) as a U.S. government-created work, it is in the public domain and; 2) if anyone were to own the rights to the video, it would be NASA, as the creator. In this case, and a practice that is not completely unusual, news channels uploaded their own news footage, which incorporated NASA’s video, and was subsequently flagged by the automated content ID systems.

Yes, the Internet provides numerous opportunities for large-scale privacy, but turning to automatic filters causes the pendulum to swing to far in the direction of rightholders, ignoring the user rights codified in important limitations and exceptions, like fair use.

Krista L. Cox is a policy attorney who has spent her career working for non-profit organizations and associations. She has expertise in copyright, patent, and intellectual property enforcement law, as well as international trade. She currently works for a non-profit member association advocating for balanced copyright. You can reach her at [email protected].

Sponsored

Topics

, , , ,