In a surprise to few people, the D.C. Circuit Court of Appeals unanimously affirmed the District Court’s decision that the Copyright Office had not abused its discretion in denying registration to the computer program / AI engine “The Creativity Machine” for the artwork “A Recent Entrance to Paradise.” Stephen Thaler, the owner, creator, and operator of The Creativity Machine, brought the suit. At argument his counsel tried to shift to an argument that Thaler was the author through his creation of The Creativity Machine, but the Court deemed that argument waived, as had the court below. Thaler’s attorney Ryan Abbot has indicated that he intends to appeal. Presumably any petition to the Supreme Court will focus on the copyrightability by AI issue and not the shift to arguing Thaler was the author, which the DC Circuit clearly disdained and the Supreme Court will almost certainly have no interest in.

The Court’s decision was focused entirely on the copyright statute and expressly avoided consideration of the question of whether the constitution permitted registration of AI works, relying on the canon of constitutional avoidance. In particular, the decision was focused on certain attributes unique to the human condition being specifically enumerated in Title 17, enumerating the ability to “have property, traditional human lifespans, family members, domiciles, nationalities, mentes reae, or signatures.” Op. at 12. This part largely mirrors the Copyright Office’s position in briefing. The Court makes clear that not all of these are indicia of inability to register a copyright, noting Belford v. Scribner and Melissa Homestead’s work on coverture and copyright, but rather that these taken together indicate that machines are excluded from copyright under the copyright statute.

The Court also noted the longstanding U.S. Copyright Office position that human authorship is required and accorded substantial weight to that position. In particular, noting the National Commission on New Technology Uses of Copyrighted Works (CONTU) commission hearings and report make this position clear, the Court believed that Congress was “adopt[ing] an agency’s interpretation of a term” with the 1976 Act, pursuant to Sackett v. EPA. I’m not sure that’s right – the 1976 Copyright Act was essentially written before the 1965 Annual Report of the U.S. Copyright Office was published, and certainly was essentially complete for years before the 1973 First Compendium. In fact, in 1973 Congress created CONTU specifically asking the Commission to inter alia “study and compile data on…the creation of new works by the application or intervention of such automatic systems or machine reproduction.” The fact that in 1978 the commission reported that the current approach was correct is challenging to impute back to 1976. The fact that Congress was concerned about the issue and unclear on the correctness of the current approach regarding human authorship seems a better takeaway than saying the approach at the time was simply incorporated into the 1976 Act. At a minimum it doesn’t seem that the issue was “well-settled” in 1976.

The Court had an interesting line about the purposes of copyright, falling heavily on the incentive theory view of copyright law, noting that copyright “is not extended as ‘a special reward’ to the author, but rather ‘to encourage the production of works that others might reproduce more cheaply.'” Op. at 3. This is followed by the famous quote from U.S. v. Loews that “[c]opyright law “‘makes reward to the owner a secondary consideration.'”

I also think the Court’s gloss on the Copyright Office’s standard for copyrightability is interesting, namely that “[t]he rule requires only that the author of that work be a human being—the person who created, operated, or used artificial intelligence—and not the machine itself.” Nothing radical, but I suspect the line will be quoted down the road.

Finally, the Court’s shout-out to Star Trek: The Next Generation – more specifically Data’s Ode to Spot – is getting a little attention, to wit:

Humans at some point might produce creative non-humans capable of responding to economic incentives. Science fiction is replete with examples of creative machines that far exceed the capacities of current generative artificial intelligence. For example, Star Trek’s Data might be worse than ChatGPT at writing poetry, but Data’s intelligence is comparable to that of a human being. See Star Trek: The Next Generation: Schism (Paramount television broadcast Oct. 19, 1992) (“Felis catus is your taxonomic nomenclature, an endothermic quadruped, carnivorous by nature”). There will be time enough for Congress and the Copyright Office to tackle those issues when they arise.

Let me know what you think led to this and whether the Trek fan is the Judge or a clerk. or both!

At the oral argument before the DC Circuit this September in Thaler v. Perlmutter, regarding whether a piece of graphical art conjured by a computer program without prompting can be registered for copyright, Judge Robert L. Wilkins asked a sensitive question of the government’s counsel – “Do you know whether enslaved people were granted copyright?” (at around 21 minutes in the audio recording).[1] I was at the argument, and the government’s counsel did not have an answer to the question. I had some suspicions, and decided to research the issue a bit.

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 is currently involved with the question of how to handle applications to register works which are partially or fully made using generative AI. A study is currently ongoing on numerous areas of AI and copyright, and over 10,000 comments have been submitted. However, aside from a isolated blog post, so far as I can tell none of them mention the “Rule of Doubt” as a way to handle these registrations (edit to add: Edward Lee posted this draft article quite recently making a similar argument). The rule of doubt nowadays is essentially only used for computer software where only object code has been deposited, but it has historically been broader. I think it offers at least a temporary solution as we figure out how copyright looks in the mid-21st century.

As I explained in Examining Copyright, the rule of doubt is both a examination doctrine and a registration caveat of substantial vintage. As the name suggests, the rule of doubt is essentially a registration with a cloud over it – the Office accepts the registration but the caveat of the rule of doubt robs it of the presumption of validity which generally attaches to a registration. Probably the best statement of the vintage rule of doubt comes from a 1969 article , which describes it as a “rule of reasonable doubt,” whereby

a claim should not be disapproved if there is a reasonable doubt as to its validity (as may ultimately be determined by a court), even though the Office itself may consider the claim invalid-hence the policy of refusing only those claims which are defective on their face.

David Goldberg & Richard Dannay, Fraud on the Copyright Office: Its Use and Misuse as a Defense in Copyright Infringement Actions, 44 N.Y.U. L. REV. 540, 557 (1969)

Although sources trace it to the 1940s, it actually goes back even earlier. In 1911 the Copyright Office accepted registrations of piano rolls as musical arrangements under a early version of the rule of doubt, even though Congress had declined to extend copyright protection to piano rolls only two years earlier. Examining Copyright at 514. Following the fee increase of 1928, owners of content for newspaper syndication began sending books of collected content instead of paying a fee per piece of content, which the Office accepted under the rule of doubt for years until they stopped doing so in the mid-1930s, leading to litigation. Id. at 519. Patry on Copyright traces the first appearance of the rule to 1942, in Assistant Register of Copyrights Herbert A. Howell’s work The Copyright Law. § 17:129. Rule-of-doubt registrations. According to the Copyright Office’s 1950 Manual, they would “register material which we feel a court might reasonably hold to be copyrightable, even though, personally, we feel that it is not subject to copyright.” In the failed Copyright Reform Act of 1993 it was proposed at Section 5(a) to codify something similar to the rule of doubt, although in the House Report (at 18) it was stated that this was not quite the same as the traditional rule of doubt.

The year 2024 marks sixty years since what is generally considered the first computer program registration – in 2019 I wrote here about how these registrations were made under the rule of doubt in 1964, and the precedent was important. With that decision, the Copyright Office managed to avoid making itself the nexus of a substantial policy debate without resolving the question conclusively, while at the same time serving its function as the administrative center of American copyright law.

The litigation over the copyrightability of typefaces in the 1970s gives further evidence of what the rule of doubt meant prior to 1978 – that the Copyright Office “will resolve any reasonable doubt concerning the copyright ability of a particular work in favor of registration.” Eltra Corp. v. Ringer, No. CIV.A. 76-264-A, 1976 WL 21070, at *3 (E.D. Va. Oct. 26, 1976), aff’d, 579 F.2d 294 (4th Cir. 1978). However, the passage of the 1976 Copyright Act changed the dynamic for the rule of doubt – previously without a registration a suit could not be brought, thus increasing the demand on the Copyright Office to approve the registration. That was no longer the case, but although consideration was given to abandoning the rule of doubt, the Copyright Office did not do so.

In 1983, the Copyright Office proposed that they would allow registration under the “‘rule of doubt’ based on an object code format as the deposit” or in cases of trade secrecy where the examiner had not seen the work. Notice of Inquiry Deposit Of Computer Programs and Other Works Containing Trade Secrets, 48 FR 22951-01 (May 23, 1983). Nowadays, “as a general rule, the Office will apply the Rule of Doubt only in” attempts to register computer software where source code is not deposited, or in situations where the examiner is not sent the deposit material due to trade secrecy. There is an allowance of other “exceptional circumstances” where it could be used, but the Copyright Office “will not register a claim under the Rule of Doubt simply because there is some uncertainty as to how that issue may be decided by a particular court.” Compendium § 607.

As indicated, section 607 of the Compendium of Copyright Office Practices contains the rule of doubt nowadays. The statement of the rule is quite broad, but it is subject to the above caveats:

On occasion, the Office may register a claim to copyright, even though the Office has reasonable doubt as to whether the material submitted for registration constitutes copyrightable subject matter or whether the other legal and formal requirements of the statute have been met. This practice is known as the Rule of Doubt.

Compendium Section 607 – Registration Made Under the Rule of Doubt

Nonetheless, there is an allowance of other “exceptional circumstances” where it could be used, but the Copyright Office “will not register a claim under the Rule of Doubt simply because there is some uncertainty as to how that issue may be decided by a particular court.” In practice, though, the rule of doubt is now limited to “registrations submitted in formats that are not understandable to humans.” Compaq Computer Corp. v. Procom Tech., Inc., 908 F. Supp. 1409, 1415 (S.D. Tex. 1995). The last registration I could find using the rule of doubt in a more traditional situation was made in 2011 – VA0001789386, Designs by Pauline Landers Circa 1945.

It may be that the traditional rule of doubt is no longer applicable under the 1976 Act, but the shift to using it only for object code and trade secrets is much more recent than that. More to the point, the use of the rule of doubt for computer software when the statutory and constitutional bases for protection were not established offers a precedent we can learn from. I don’t claim to know if works which incorporate generative AI are copyrightable, and I’m in fact fairly skeptical of claims that they are at least as to the AI portions. But the administrative structure of the Copyright Office is a difficult place to have this discussion, as a recent article in Wired shows. As the Copyright Office begins to offer more flexibility on AI, the rule of doubt is a way to do so without granting prima facie validity if Courts find that copyright does not cover these works.

On November 17, 2015, Robert Brauneis delivered the 12th Christopher A. Meyer Memorial Lecture on “The Landscape of American Cultural Production, 1978-2012: A Quantitative Look Through the Lens of Copyright Registrations.” The lecture is an annual series from the Copyright Society of the USA, along with the U.S. Copyright Office and others. The video of the lecture has long been online (although I can’t embed it here), and the second half of the talk became An Empirical Study of the Race, Ethnicity, Gender, and Age of Copyright Registrants, published in 2018 with Dotan Oliar. The raw data and online appendix are also worth looking at. However, an earlier part of the talk, dealing with the number of individual and corporate registrants, has not been published. Bob was good enough to send me his slides, and permission to share them. To be clear, beyond the interstitial text none of this is my work, and also this data is now about a decade out of date and would need a re-do for further analysis.

In the 1890s, famed dancer Loïe Fuller had her first major success with the “Serpentine Dance,” a variation of the popular “skirt dances” of the day. But should she keep it for herself? When others started dancing an imitation of her choreography she would try, and in the process give chreographic copyright a false start a half century before it would succeed.

The case of Fuller v. Bemis is fairly well-known to historians of copyright and dance, and there’s already been an excellent post about it on IPKat. Anthea Kraut has also written both an article and a book that focuses, in full or part, on this case. In that case the Fuller sued dancer Minnie Renwood Bemis for copying her “Serpentine Dance” in 1892. However, the Southern District of New York dismissed her claim and found any copyright in the dance invalid.

“A stage dance illustrating the poetry of motion by a series of graceful movements, combined with an attractive arrangement of drapery, lights and shadows, but telling no story, portraying no character and depicting no emotion, is not a ‘dramatic composition’ within the meaning of the Copyright Act.”

Fuller v. Bemis, 50 F. 926, 929 (C.C.S.D.N.Y. 1892)

However, had Fuller received a registration? No registration is indicated in the Virtual Card Catalog.

While Southern Illinois – and its concomitant U.S. District Court – isn’t necessarily a hotbed of copyright litigation, that changed last week. In 2018, Catherine Alexander sued a number of video game companies as well as World Wrestling Entertainment (WWE) for copyright infringement in the Southern District of Illinois (docket here). Years earlier, she had done extensive tattoos on the wrestler Randy Orton, who subsequently found great success in the WWE. He was included in numerous iterations of the “WWE 2K” videogame line, with his tattoos faithfully replicated.

In September of 2020, the court denied summary judgment, holding triable issues of fact regarding whether the use was fair and whether an implied license existed. The trial was held at the end of September, and at the end of the week the jury returned a verdict for Alexander – but the victory was short lived as damages of only $3,750 were awarded. In the end the case left no-one satisfied – the game publisher and WWE still have difficult clearance issues, and the recovery surely was not what Alexander was hoping for. Experts have also criticized the jury instructions for the failure to include a question of implied license even though both sides asked for it – and indeed that was the main question the jury should have been answering.

SIU Law 2L Taylor Ingram was there for the second day of the trial (Tuesday, September 27, 2022), and recorded some impressions of the case, which I’ve shared below. I think they shed some light on the case and the real people involved – her observations and thoughts follow below: