The United States Copyright Office released its second report dealing with artificial intelligence (AI) and copyright. The report, Copyright and Artificial Intelligence Part 2: Copyrightability, unsurprising, concludes that copyright does not extend to purely AI-generated material where there is insufficient human control over the expressive elements. However, and consistent with copyright law, it concluded that copyright can subsist in generative AI outputs that contain copies of inputs that are perceptible in AI-generated outputs, as well as where persons take AI-generated outputs and select, arrange, coordinate, or modify them to meet the threshold requirements of originality. However, the Copyright Office concluded that based on the functioning of current generally available technology, prompts do not alone provide sufficient attributes of human authorship for copyright to subsist in the outputs. For the reasons explained below, this latter conclusion is too broadly stated and makes unsupportable assumptions.
Copyright in computer aided versus computer generated works
It is a fundamental principle of copyright law that copyright can only subsist in works that contain the minimal degree of originality under applicable copyright laws. In the U.S. for a work to be “original” it must not be copied from another work and must meet the minimal “creativity” standard.[i] In Canada, “skill and judgement” is required for a work to be original.[ii] In either case, the only originality that qualifies is human authorship. Content that is generated using computer aided tools can meet that standard, but content that is generated using tools that are responsible for all of the content does not qualify.
It is sometimes difficult to draw the line as to when copyright will subsist in a work that is created using sophisticated tools. Cases around the world have canvassed this question and the determination of whether copyright can subsist is based on an analysis of the extent to which the expression in the content created was directed, controlled, or fashioned by the person using the tool. I summarized this test in my book Sookman Computer, Internet, and e-Commerce Law this way:
The determination as to whether a person is an author of a work created using a computer program will be a factual one. If the person controlling the program can be seen as directing or fashioning the material form of the work, it will likely be protected as a computer‑assisted work. But, if the person does not contribute sufficient independent intellectual effort to satisfy the Act’s requirement for originality, the individual using the program would likely not be considered its author.
Generative AI outputs that contain copyrighted inputs
The Copyright Office confirmed that if generative AI systems take inputs from users in the form of text, images, audio, video, or a combination of these mediums and use these as the basis for an intended output, the resulting output may have copyright protection, at least to the extent it continues to reflect expression in the original input. The report contains the following example of a work which would qualify for protection.
Modifying or Arranging AI-Generated Content
Copyright law can also protect the selection or arrangement of unprotectable subject matter – like generative AI content – or content that has been modified to the extent that the modifications reflect a sufficient amount of original contribution.
The Copyright Office confirmed that generating “content with AI is often an initial or intermediate step, and human authorship may be added in the final product”. As explained, quoting from the Copyright Office’s AI Registration Guidance, “a human may select or arrange AI-generated material in a sufficiently creative way that the resulting work as a whole constitutes an original work of authorship.” “A human may also “modify material originally generated by AI technology to such a degree that the modifications meet the standard for copyright protection.”
The Report gives the following illustrations of these principles.
Copyright in outputs generated using prompts.
The Copyright Office concluded that “given current generally available technology, prompts alone do not provide sufficient human control to make users of an AI system the authors of the output”. The reason, according to the Copyright Office, is that “[p]rompts essentially function as instructions that convey unprotectible ideas. While highly detailed prompts could contain the user’s desired expressive elements, at present they do not control how the AI system processes them in generating the output”.
The Office’s rational that prompts cannot determine expressive elements was further explained as follows:
In theory, AI systems could someday allow users to exert so much control over how their expression is reflected in an output that the system’s contribution would become rote or mechanical. The evidence as to the operation of today’s AI systems indicates that this is not currently the case. Prompts do not appear to adequately determine the expressive elements produced, or control how the system translates them into an output. The gaps between prompts and resulting outputs demonstrate that the user lacks control over the conversion of their ideas into fixed expression, and the system is largely responsible for determining the expressive elements in the output. In other words, prompts may reflect a user’s mental conception or idea, but they do not control the way that idea is expressed. This is even clearer in the case of generative AI systems that modify or rewrite prompts internally.
The Office also considered even iterative prompting to be insufficient to produce a copyrighted work because “revising and submitting prompts multiple times, the user is “re-rolling” the dice, causing the system to generate more outputs from which to select, but not altering the degree of control over the process. No matter how many times a prompt is revised and resubmitted, the final output reflects the user’s acceptance of the AI system’s interpretation, rather than authorship of the expression it contains.”
In my view, the Copyright Office’s categorical rejection of any possibility that prompting could give rise to some content that exhibits a sufficient degree of human authorship suffers from several shortcomings – including failures to consider some of the nuances of U.S. copyright law and its assumption that iterative promoting can never result in human authorship.
First, it is well established under U.S. copyright law that although copyright cannot subsist in mere ideas, copyright can subsist in a combination, association, or arrangement of ideas or facts in such a way that the resulting work as a whole constitutes an original work of authorship.[iii] Thus, the Copyright Office’s assumption that prompts can only contain unprotectable ideas is too broadly stated. Further, if an original combination of ideas is replicated in computer generated outputs, it is possible that the prompts would be protected even if embodied in expression that otherwise would be unprotectable. Thus, the Copyright Office’s conclusion that “prompts may reflect a user’s mental conception or idea, but they do not control the way that idea is expressed”, is also too broadly stated. It ignores that the copying of the combination of ideas into the outputs would be protectable irrespective of whether other “expression” in the output would also be protected.
Second, it is also well established under U.S. copyright law that a work can be infringed without appropriating the literal expression in the work. Under the doctrine of “comprehensive non-literal similarity”, for example, copyright can protect against the copying of non-literal expression including where there is no word-for-word or literal similarity but where the defendant has nonetheless appropriated the “fundamental essence or structure” of plaintiff’s work. A plaintiff can succeed in an infringement suit under this doctrine by showing, for example, that the plot, themes, dialogue, mood, setting, pace, and sequence in a literary work has been appropriated.[iv] Here again, if a person creates detailed prompts that contains the fundamental essence or structure for a work, or a sufficiently detailed description of the desired plot or other creative elements for a work, computer generated output that contains these original elements within wholly computer generated expression, could also be protectable.[v]
Of course, not all ideas or combinations of ideas used in prompts or contained in computer generated outputs would be protected by copyright. That is because copyright does not protect mere ideas or concepts and various doctrines such as merger and scenes a fair, would preclude protection in both the inputs and outputs. Further, the copyright in the outputs would necessarily be thin and be limited to only the copyrightable elements contained in the prompts. In this respect, the outputs could be protected much like the graphic shown above where the generative AI outputs are protected because, and only to the extent that, they copy the copyrighted inputs.
While the Copyright Office made references to “prompt engineering”, the “practice of crafting prompts that are optimized to elicit a desired result”, it appears to have summarily rejected these advancing techniques because “the output of current generative AI systems may include content that was not specified and exclude content that was” and because of uncertainty and unpredictability around how a particular prompt or other input will influence the output. As noted above, it also rejected that iterative prompts could result in original content because that would still lead to unpredictable results. These reasons are not compelling for several reasons.
First, it is hard to see why a copyrightable work could not come into existence merely because outputs may contain some content not specified or exclude specified content. Copyright does not need to extend to all parts of a work for copyright to subsist in it.
Second, the Report did not analyze the different prompt engineering techniques or contemplate future developments that could provide persons with a degree of control over the outputs produced using a generative AI tool. The Report assumed that the degree of control would only have to emerge from developments in AI systems. If advanced prompt engineering tools use AI to generate prompts, further questions would arise as to the copyrightability of prompts generated using these tools.
Third, the Copyright Office’s assumption that prompts may yield unpredictable results does not logically support the inference that iterative prompting including iterative rejection of outputs that do not produce the desired results can never result in a sufficient level of control to demonstrate human authorship. The protection for such outputs may be thin, but the “skill and judgement” or “creativity” to create the outputs may well meet the relatively low threshold to meet the “originality” standards under applicable law. There currently is litigation before the U.S. courts that may prove the Copyright Office wrong on this point.
For more on this, topic see
- Barry Sookman, Copyright and Generative AI: Understanding Recent Chinese Court Decisions
- Barry Sookman, Do generative AI inventions and works qualify for patents and copyrights? The Thaler and SURYAST decisions
- Lee, Edward, Comment of Professor Edward Lee to Artificial Intelligence Study by The United States Copyright Office (October 30, 2023). Available at SSRN: https://ssrn.com/abstract=4619118or http://dx.doi.org/10.2139/ssrn.4619118
[i] Feist Publ’ns, Inc. v. Rural Tel. Serv. Co., 499 U.S. 340, (1991)
[ii] CCH Canadian Ltd. v. Law Society of Upper Canada, 2004 SCC 13
[iii] Gordon v. Weir, 111 F. Supp. 117(E.D.Mich, 1953), aff’d, 216 F.2d 508 (6 Cir. 1954); Nash v. CBS, Inc., 691 F. Supp. 140 (N.D. Ill. 1988); Gaiman v. McFarlane, 360 F.3d 644 (7th Cir. 2004); Corwin v. Walt Disney Co., 475 F.3d 1239, (11th Cir.2007).
[iv] Arica Institute, Inc. v. Palmer, 970 F.2d 1067 (2d Cir. 1992); Shaw v. Lindheim, 908 F.2d 531 (9th Cir. 1990).
[v] The law in this area is very similar in Canada, as summarized by the Supreme Court of Canada in Cinar Corporation v. Robinson, 2013 SCC 73, summarized in my blog post here.
The Act protects original literary, dramatic, musical, and artistic works: s. 5. It protects the expression of ideas in these works, rather than ideas in and of themselves: CCH Canadian Ltd. v. Law Society of Upper Canada, 2004 SCC 13…
A substantial part of a work is a flexible notion. It is a matter of fact and degree. “Whether a part is substantial must be decided by its quality rather than its quantity”: Ladbroke (Football), Ltd. v. William Hill (Football), Ltd., [1964] 1 All E.R. 465 (H.L.), at p. 481, per Lord Pearce. What constitutes a substantial part is determined in relation to the originality of the work that warrants the protection of the Copyright Act. As a general proposition, a substantial part of a work is a part of the work that represents a substantial portion of the author’s skill and judgment expressed therein.
A substantial part of a work is not limited to the words on the page or the brushstrokes on the canvas. The Act protects authors against both literal and non-literal copying, so long as the copied material forms a substantial part of the infringed work. As the House of Lords put it,
the “part” which is regarded as substantial can be a feature or combination of features of the work, abstracted from it rather than forming a discrete part. . . . [T]he original elements in the plot of a play or novel may be a substantial part, so that copyright may be infringed by a work which does not reproduce a single sentence of the original. (Designers Guild Ltd. v. Russell Williams (Textiles) Ltd., [2001] 1 All E.R. 700, at p. 706, per Lord Hoffmann; see also Nichols v. Universal Pictures Corporation, 45 F.2d 119 (2d Cir. 1930), per Learned Hand J.)
