As generative AI technologies such as GPT-4 and Midjourney have rapidly become more sophisticated and their creative use has exploded in popularity, the US Copyright Office today released guidelines to clarify when material generated by IA may be copyrighted.
The advice comes after the Copyright Office ruled that an author could not protect individual AI images used to illustrate a comic because each image was generated by Midjourney, not a human artist. . In making its decision, the Copyright Office pledged to uphold the long-standing legal definition that authors of creative works must be human to register works. For this reason, officials have confirmed that AI technologies can never be considered perpetrators.
It wasn’t the only case to influence the new guidelines, but it was the most recent. The struggle with the complex issues of comic book authorship prompted the Office of Copyright to launch an agency-wide initiative to continue to explore a wider range of copyright issues arising as the AI models used to generate text, art, audio, and video continue to evolve.
The guidance offers some details on what is not eligible for copyright when dealing with AI works generated only by prompts – without any editing – which the Copyright Office equates to give “instructions to a commissioned artist”. These works do not have human authorship and, therefore, will not be registered.
When “AI technology receives only a prompt from a human and produces complex written, visual, or musical works in response, the ‘traditional elements of authorship’ are determined and executed by the technology, not by the human user,” the guide explains. “Based on the Bureau’s understanding of currently available generative AI technologies, users do not exercise ultimate creative control over how these systems interpret prompts and generate material.”
However, as in the case of Midjourney, an author who arranges generative AI in a specific sequence – like the layout design of a comic book – can potentially protect that sequence of images, if the arrangement is “creative enough”. Similar logic applies if an author or artist has modified AI-generated material and “the modifications meet the standard of copyright protection”. Examples could be editing an AI image in Adobe Photoshop or editing AI-generated audio using guitar pedals, the guide says.
It is clear, however, that the Copyright Office is only in the early stages of navigating these complex cases, and the guidance remains somewhat vague. Ultimately, officials who question whether AI-assisted works were designed by humans or machines will make decisions on a case-by-case basis, according to the guidelines.
“The answer will depend on the circumstances, in particular how the AI tool works and how it was used to create the final work,” the guide says.
Any AI-generated content must be disclosed
Perhaps the most important aspect of the guidelines is an author’s “duty to disclose the inclusion of AI-generated content in a work submitted for registration”.
When registering works, authors must distinguish between what content is human-made and what content is AI-generated. If applicants are unsure how to reference AI-generated content, the Copyright Office recommends providing a general statement that the work contains AI-generated content. This will prompt the office to follow up to help each author fill in the blanks in an application.
For artists who have pending applications or who have already registered works containing AI-generated content, the Copyright Office suggests correcting the public record by submitting an additional registration. Any failure to accurately reflect the role of AI in copyrighted works could lead to the “loss of the benefits of registration”, the office warned. This could leave works vulnerable to copying, with little or no legal recourse for copyright infringement claims.
Failure to disclose AI-generated content is the only type of offense addressed in the guide. Critics like Alex J. Champandard, co-founder of Creative.ai, a group of hackers and artists interested in generative AI,tweeted say the current guidelines put writers in a precarious wrestling-22 position.
“By disclosing the AI, you expose yourself to a violation, but by not disclosing the AI, it is safer but in violation of [the US Copyright Office]!” Champandard’s tweet suggested as much.
Ars could not reach Champandard to discuss his group’s other concerns about the tips.
More advice to come in 2023
The Copyright Office knows it still has a lot of work to do to fully clarify when AI-assisted content can be registered. To that end, it has scheduled a series of listening sessions in April and May to gather feedback from the public. The office has also launched a dedicated webpage for posting updates on AI-related news and events, so it’s easier to follow the advice as the rules are updated.
On April 19, the listening sessions kick off with an event dedicated to AI-generated literary works. This is followed by a session on visual arts on May 2, audiovisual works on May 17 and music and sound recordings on May 31.
Artists, creative industry players, AI developers, AI researchers and lawyers are all encouraged to register to attend these sessions. But because attendance will necessarily be limited, the Copyright Office has also indicated in guidelines released today that there will be additional opportunities for stakeholders to engage beyond attendance at these sessions.
“The Bureau intends to issue a Notice of Inquiry later this year seeking public comment on other legal and policy matters, including how the law should apply to the use of copyrighted works in AI training and the resulting processing of results,” the guide states.
It will likely be a daunting task to move the guidance forward, but Nora Scheland, public affairs specialist for the US Copyright Office, told Ars the office is “excited” about launching a broader initiative exploring the IA and “will be looking forward to feedback and feedback from the public.” throughout the months to come. »