What Does John Cage Have to Do with AI Authorship?

As some of you know I’m working on a project on the history of copyright protection for computer-generated works, aka AI authorship. As that piece slowly moves ahead I wanted to share one of the more surprising things I’ve discovered which seems relevant – the history of copyright registration for musical works utilizing chance and randomness. To cut to the chase, in the 1960s the U.S. Copyright Office considered registrations of a number of such works, and determined that in cases where there was not sufficient musical authorship – most famously in the “silent piece” 4’33”, registration could be made of the instructions as a textual work. This seems like it could operate as a precedent for registering the prompts fed into generative AI like ChatGPT even if the output is not registrable at the U.S. Copyright Office.

The Copyright office has made clear, going back to 1965, that works where a computer was responsible for the creative component of a work cannot be registered for copyright. There are a number of recent situations involving attempts to register works for copyright where the Copyright Office has partially or wholly rejected the work, including a recent lawsuit filed in Colorado. There are many other cases, going back in history into the 1950s, of computer-generated works being sent to the U.S. Copyright Office for registration. The first such example, which was apparently rejected, was for the song “Push Button Bertha” where a computer wrote the music, although there is no official source for this.

While a lot has been written before about the precursors to AI (and I’m writing more on the subject myself), very little has been written about copyrightability of music where the authorship does not come from the composer through other means – methods such as randomness, chance, and the purposeful absence of standard musical composition. Perhaps strangely, this is an area where there is a substantial amount of precedent, and it seems worth sharing.

The idea of chance in making music is nothing new, the dice game attributed to Mozart is only one of many examples through history. In the early 1950s the composer John Cage pushed this much farther into what would come to be called aleatory music, using randomness system of the I Ching to compose such works dictated by chance as Imaginary Landscape No. 4 for 12 radio receivers, and Music of Changes for piano. In 1952 this led to his best-known and most controversial creation: 4′33″. The “silent piece,” contains no musical notation beyond three movements and an instruction to silent and although the piece’s name refers to four minutes and 33 seconds (or 273 seconds, since absolute zero is -273 Celsius), it may be performed for any length of time on any instrument.

Of course, we’re here to talk about how such a work would be registered for copyright. Cage’s publisher Henmar Press applied to register many of them – including 4’33” – with the U.S. Copyright Office, and in looking at how the applications were handled (you can see them all here) there’s some valuable lessons to learn about registration of works of indeterminate authorship. Most notably, although a decent number of Cage’s compositions passed muster and were registered as music, others including 4’33” did not, and after years of considering the issue, were registered instead as textual works by the U.S. Copyright Office.

The registrations (the textual registrations were scanned by the Copyright Office as part of their project of scanning record books, the musical registrations are my scans) show that for works like 4’33”, although the Copyright Office received the deposit in 1965, the application was not received until 1968, leading me to surmise that initially a registration for a musical work had been submitted, it was rejected, and on the advice of the Copyright Office a replacement application to register 4’33” as a textual work was submitted and accepted.

For once, though, I didn’t need to surmise – Dorothy Pennington Keziah, the head of the music section of the Examining Division of the Copyright Office, published an article explaining the whole thing. In the passage below, Keziah is in fact describing exactly the process of deciding to register certain of Cage’s works as text rather than music.

[W]hile both the original creative impulse (concept) and the final effect (performance) may be musical in nature, the creative authorship (expression) that is fixed is not musical. It is not musical in the sense that it does not fix in tangible form individual sounds that are the result of original creative thought. Consequently, since it is the sound and not the notations themselves that constitute the subject matter of a musical copyright, and since indeterminate works such as these do not contain fixed sounds that are the result of original creative thought, it does not appear that such works would qualify for registration in Class E as musical compositions
…
Certainly, any borderline cases should be resolved in favor of registration in Class E. However, it seems clear that where the textual description does not sufficiently fix at least a minimal amount of the original sounds in the work, registration would more appropriately be made, if the work were published with notice, in Class A as a “book.”

Dorothy Pennington Keziah, Copyright Registration for Aleatory and Indeterminate
Musical Compositions, 17 Bull. Copyright Soc’y U.S.A. 311, 354 (1970).

So, this is all fascinating. But what does it have to do with AI? Simply put, where a machine cannot be an author, and the question is whether or not the author or composer has created a work of authorship protectible as a “writing” under the Constitution and statutes, aleatory music is I would argue essentially the same thing as a work created by AI following a series of prompts. After all, consider the Copyright Office’s statement about copyrightability of AI works from 1965 – the very same year applications to register Cage’s aleatory works were submitted:

The crucial question appears to be whether the “work” is basically one of human authorship, with the computer merely being an assisting instrument, or whether the traditional elements of authorship in the work were actually conceived and executed not by man but by a machine.

1965 Annual Report of the Register of Copyrights at 5

Indeed, it seems clear that essentially the same standard was being used regarding Cage’s works – was the musical expression his work, or was he simply providing a venue of sorts to hear what was already there. Even more interesting, then, is the decision that was made as to registration, which was not a flat denial. Rather, the Copyright Office looked to see what could be registered.

As such, an approach to registration of works created at least in part with generative AI would seem to follow that the Office would first look to see if there was the necessary quantum of human authorship on the part of the author, and if not could consider registration of the prompts given to the generative AI as textual works on a sort of “thin copyright” theory.

So far as I can tell this has received no attention thus far. I tweeted about this when I first discovered it in August and Ed Lee cites that tweet in his new article Prompting Progress (which is well worth reading). There has been some scholarly discussion in 2005 and 2013 into copyright for Cage’s aleatory works more generally, although it is very focused on the United Kingdom.

Much more to come on this, and I wanted to get it out as conversations about generative AI and copyright law continue to swirl. Obviously there’s a whole other conversation about training and outputs as infringing the work of creators as well.