Like many Americans, Adafruit decided to celebrate the recent 4th of July with some patriotic music.  They scored a video celebrating the 4th of July and US manufacture of Arduino with America the Beautiful, as performed by the United States Navy Band (a work on the public domain – more on that below).  Soon thereafter, they received a note from YouTube telling them that someone had issued a copyright claim against the video.  After some outrage, the claim was dropped.  But what actually happened here?  And why did it happen?

The Song is in the Public Domain

Let’s start with the song. Musical copyright is super annoying (for a great dive into one small part of music-related copyright – the ability for artists to reclaim rights they gave to labels, check out this whitepaper by my former PK colleague Jodie Griffin), but this instance is fairly straightforward.  A recorded song has two copyrights: the copyright in the underlying work (the song as written) and the copyright in the performance of that work.  Let’s take them one at a time.

The underlying work is America the Beautiful.  Wikipedia is as good a source as anything for this, and it says that the music for America the Beautiful was originally written in 1882, the poem that became the lyrics was originally written in 1895, and the two were combined in 1910.  Using the copyright duration/public domain calculator of your choice, we can be confident that works originally published on those dates are in the public domain (I’m leaving out a possible newer arrangement of the music for this analysis because no one has suggested it was a factor and it complicates things). Copyright one: check.

The performance is the recording of the actual band performing the song.  It is pretty likely that this performance happened at a time where, if it was eligible for copyright protection, it would still be protected by copyright.   A good public domain rule of thumb is anything after 1922 is still protected by copyright, and the technical realities of 1920s-era recording means any high quality recording is much newer than that.

However, there is another factor to consider.  As I am admittedly super aggressive about mentioning, works by the US government are not eligible for copyright protection.  They are in the public domain from the moment they are created. Since the US Navy Band is part of the US government…. Copyright two: check.  Former Public Knowledge intern Ethan James covered this territory about this time last year.

So What Happened?

If the song (composition and performance) are in the public domain, how could someone have asserted a copyright claim against it? Enter YouTube’s Content ID system.  Content ID is a private alternative to the DMCA that YouTube has set up on its site.  Large content owners upload music or movies that they own.  Whenever anyone else uploads a video to YouTube, YouTube checks the video against this blacklist to see if there is a match.  If there is a match, the rightsholder gets to decide what happens. Sometimes the video is taken down, sometimes the rightsholder runs ads against it, and sometimes the rightsholder prevents the uploader from running ads against the video.

There are a number of legitimate criticisms against Content ID, many of which focus on the fact that it lacks many of the minimal protections for misidentification built into the DMCA.  To continue a theme, former Public Knowledge Artist in Residency Elisa Kreisinger’s Fair Use® explores some of these shortcomings (Public Knowledge is on this stuff people).

In this case, the copyright flag was tied to Rumblefish, a service that (among other things, I assume) patrols YouTube for copyright infringement on behalf of its clients. While the Navy Band’s performance was in the  public domain, it sounds like it was close enough to the (copyright protected) performance by another band represented by Rumblefish to be flagged.

As an aside, even if this other band’s performance was identical to the Navy Band’s, it could still be protected by copyright.  Copyright allows for independent creation.  As long as the unnamed other band wasn’t actually copying the Navy Band – but rather getting to the same place for the same reasons – the Navy Band’s performance could be in the public domain while the other band’s performance could be protected by copyright.  That’s why you have to pay for the London Philharmonic’s recording of fancy classical music but can also get a copy for free from Musopen (Public Knowledge connection: they were on the podcast in 2012).

So YouTube’s Content ID confused the Navy Band’s America the Beautiful with Rumblefish’s client’s America the Beautiful and handed Rumblefish a veto over Adafruit’s video.  Because Adafruit is well known, they were able to raise a stink and have Rumblefish back down. This is a win(ish) for Adafruit, but being big enough to raise a stink really shouldn’t be a prerequisite for overcoming bad copyright claims.

What Could be Done Better/Where Should I Direct My Rage?

Content ID is imperfect. That’s inevitable, but knowing that Google and YouTube could work harder to build safeguards into the system.  And knowing that it represents performances that are dangerously similar to public domain ones, Rumblefish could work harder to tread lightly.

In this case, one obvious change would be to how Rumblefish responded to the match.  From what I can tell, Rumblefish had a match to this performance set to “prevent monetization by the uploader.”  While that is better than “take down forever,” it still had the effect of dictating terms for a video it had no rights over.  For performances of songs like America the Beautiful – songs that are in the public domain, that are likely to have multiple similar versions, and that are likely to have public domain and/or freely licensed versions floating around - Rumblefish could set a match to “flag for me but do nothing until an actual human being reviews the match.”  That is doubly true because it is much more likely that the free, public domain version of the song is incorporated into an uploaded video than the one represented by Rumblefish.

Is that even an option in Content ID?  I don’t know. If it is, Rumblefish should take it. If it isn’t YouTube should implement it.  The cost to free expression of not having that seems a bit too high not to.

The other thing that YouTube could do would be to build some sort of penalty for false positives into the system.  An occasional incorrect match is one thing, but if a song that is included in Content ID regularly triggers misidentification perhaps it should not be allowed to claim control over videos.  This kind of common sense restriction could go a long way towards avoiding these types of situations.

Licenses are Not Proxies for Openness in AI Models

Earlier this year, the National Telecommunications and Information Administration (NTIA) requested comment on a [number of questions](htt...… Continue reading