The Copyright Office Issues Guidance Regarding Works Produced by Generative AI

By Michael Borella --

On March 16, the Copyright Office published guidance in the Federal Register relating to works produced at least in part by generative artificial intelligence (AI).  This is the latest in a series of policy decisions and statements that the Office has made to applicants attempting to register such works.

While AI has been a viable technology for decades, improvements in computer processing and storage capabilities of the last 15 years have enabled the rise of machine learning, a branch of AI in which a computer system can be trained on a large amount of data in order to "learn" underlying patterns within.  In many cases, these patterns are too subtle for a human to notice or require analysis of a massive data set in order to be perceived.  As an example, spam filters often use trained machine learning models to classify email messages as either spam or not spam.

But generative AI takes machine learning to a whole new level.  In 2022, sophisticated generative AI tools were released to the public for the first time.  Rather than considering an observation of data and classifying it in some fashion, generative AI models can create new observations -- particularly images (DALL-E 2, Stable Diffusion, and Midjourney) and text (ChatGPT) -- from a user's textual prompt.

The results can be simultaneously remarkable, startling, and disturbing.  But since the models were trained on existing works found on the Internet, they may exhibit some (or many) similarities with these previous works.  Thus, one issue is whether the output of generative AI is a derivative work or infringes upon the rights of a previous author.

The purpose of this guidance, however, is to inform the public about the Office's current policy regarding "whether the material [the generative AI models] produce is protected by copyright, whether works consisting of both human-authored and AI-generated material may be registered, and what information should be provided to the Office by applicants seeking to register them."

At one end of the spectrum, the Office has a long-held position that copyright can only be used to protect the products of human creativity.  Therefore, when presented in 2018 with an application for an image that was described as ''autonomously created by a computer algorithm running on a machine," the Office denied registration to the work as lacking a human author.  Last year, the Office initially registered a graphic novel in which the images were generated using Midjourney and then combined with human-authored text, but recently modified the registration to indicate "that the individual images themselves could not be protected by copyright."

The guidance adds some color to the Office's decision making process regarding the human authorship question.  Particularly, the Office states:

In the case of works containing AI-generated material, the Office will consider whether the AI contributions are the result of ''mechanical reproduction'' or instead of an author's ''own original mental conception, to which [the author] gave visible form.''  The answer will depend on the circumstances, particularly how the AI tool operates and how it was used to create the final work.  This is necessarily a case-by-case inquiry.

As an example, if the human merely provided a prompt to the generative AI model and then the model goes on to produce "complex written, visual, or musical works in response," then the ''traditional elements of authorship are determined and executed by the technology—not the human user."  In that context, the Office notes that "prompts function more like instructions to a commissioned artist" who then makes the artistic decisions.

But the Office recognizes that there can be a threshold amount of human creativity within a work that also contains the output of a generative AI model such that the work as a whole or parts thereof are registerable.  The Office provides examples such as where "a human may select or arrange AI-generated material in a sufficiently creative way" or that "an artist may modify material originally generated by AI technology to such a degree that the modifications meet the standard for copyright protection."  Nonetheless, in these hybrid cases the copyright protection will only extent to the human-authored aspects.

With these principles in place, the Office goes on to provide procedural guidance to applicants seeking to register hybrid human / AI works.  Notably, they must "identify the author(s) and provide a brief statement . . . that describes the authorship that was contributed by a human."  For a work that "creatively arranges the human and non-human content," the applicant must "describe human-authored content created by the author and describe AI content generated by artificial intelligence."  Alternatively, the applicant can "provide a general statement that a work contains AI-generated material."  Then, the "Office will contact the applicant when the claim is reviewed and determine how to proceed."

For existing applications that do not adhere to these guidelines, the Office requires that the applicants "contact the Copyright Office's Public Information Office and report that their application omitted the fact that the work contained AI-generated material."  This will be considered by the examiner when considering the application.

To correct an existing registration, the applicant must submit a supplementary registration that describes "the original material that the human author contributed" and "disclaim the AI-generated material."  The Office will issue a supplementary registration certificate if there is sufficient human authorship.

If an applicant fails to update such applications and registrations, the registrations may be cancelled.  Likewise, a court can disregard a registration if it was obtained through deceiving the Office regarding aspects of its authorship.

Generative AI has proven to be a technologically and socially disruptive paradigm.  Individuals, businesses, and governments will likely be wrestling with its implications for years to come.  In the mean time, it is good to see the Copyright Office being proactive regarding hybrid authorship rather than letting these issues be addressed piecemeal in the courts.  Like many issues in copyright, the authorship inquiry is likely to be highly fact-sensitive and subject to few clear lines of demarcation.

Comments

[skeptical]

Pardon the application of jargon, but Copyright appears to present an easier call in comparison to patents.

By this, I note that Copyright is not encumbered by the necessity of passing a type of 103 test — that is, Copyright is only concerning itself with an authorship question.

Patent law is now (even if no one is discussing it yet) at a point to which the non-human legal fiction of a Person Having Ordinary Skill In The Art presents a bar to proper grant of patents, and this bar represents a State of the Art that ever more so will include non-human inventions.

This problem is only exacerbated with the ‘trend’ of treating 103 art as mere aggregates of 102 art (witness the decline of “motivation to combine” to merely being “motivation”).