The US Copyright Office has released guidelines today to make clear when AI-generated content can be protected in light of the rapidly increasing sophistication of generative AI technologies like GPT-4 and Midjourney and the explosive growth in popularity of their creative applications.
The Copyright Office’s ruling that an author could not copyright specific AI images used to illustrate a comic book is the source of this advice, as each image was created by Midjourney rather than a human artist. The Copyright Office pledged to upholding the long-standing legal requirement that authors of creative works must be human in order to register works when it made its decision. Officials confirmed that AI technology can never be regarded as writers as a result.
Although not the only instance that affected new instructions, this was the most recent. The Copyright Office launched an agency-wide initiative to continue investigating a wider range of copyright issues developing as the AI models used to generate text, art, audio, and video continue to advance as a result of the difficult authorship issues surrounding the comic book.
When it comes to AI works produced simply by prompts—without any modifications—which the Copyright Office compares to delivering “instructions to a commissioned artist,” the guideline provides some specifics on what isn’t copyright able. These works are not human-authored, hence they cannot be registered.
Yet, just like in the Midjourney case, if an author arranges generative AI into a predetermined sequence—such as creating the layout for a comic book—and the arrangement is “sufficiently creative,” they may be able to copyright that series of images. If an author or artist modifies AI-generated content and “the adjustments fulfill the requirement for copyright protection,” then the same reasoning holds true. According to the advice, such examples include editing an AI image in Adobe Photoshop or changing AI-generated sounds with guitar effects.
The “conventional components of authorship” are chosen and carried out by the technology—not the human user—when “AI technology gets merely a cue from a person and produces sophisticated literary, visual, or musical works in response,” the guidance continues. According to the Office’s understanding of the current state of generative AI technologies, users do not have complete creative control over how these systems read instructions and produce content.
Yet it’s obvious that the Copyright Office is still at the beginning of dealing with these complicated instances, and the advice is still a little hazy. Officials will ultimately decide whether AI-assisted works were conceptualized by humans or by computers on a case-by-case basis, according to the guidance.
The guidance states that the answer will depend on the circumstances, including how the AI tool functions and how it was used to create the final product. The author’s “obligation to disclose the incorporation of AI-generated content in a work submitted for registration” is arguably the most important component of the guidance.
Authors must distinguish between AI-generated and human-authored content when registering works. The Copyright Office advises including a generic statement that the work comprises AI-generated content if applicants are unsure on how to refer to the AI-generated content. The office will then follow up to assist each author in completing the gaps in an application. The Copyright Office advises artists who have outstanding applications or who have already registered works that contain AI-generated content to submit a supplementary registration to amend the public record. Losing the advantages of the registration could occur if the function of AI in copyrighted works is not accurately reflected, the office cautioned. As a result, there may be little to no legal consequences for copyright infringement accusations, making works susceptible to copying.
The guidance solely addresses one kind of infringement: failure to disclose AI-generated content. The existing advice, according to critics like Alex J. Champandard, co-founder of Creative.ai, a group of hackers and artists interested in generative AI, puts authors in a perilous catch-22 situation. “By disclosing the AI, you’re exposing yourself to infringement; by not disclosing the AI, it’s safer but against [the US Copyright Office]!” Champandard tweeted an idea.
The Copyright Office is aware that there is still much to be clarified on when AI-assisted content can be registered. In order to get feedback from the public, it has scheduled a series of listening sessions for April and May. It is now simpler to stay informed about guidelines when regulations change thanks to the office’s introduction of a webpage devoted to posting updates on AI news and events. Listening sessions begin on April 19 with a program highlighting literary creations by artificial intelligence. Next, on May 2, an event will focus on visual arts, on May 17, on video works, and on May 31, on music and sound recordings.
Nora Scheland, a public affairs specialist for the US Copyright Office, told Ars that the office is “excited” about launching a larger initiative exploring AI and will “look forward to public input and feedback throughout the coming months.” However, developing guidance will undoubtedly be a difficult task.
“The Office expects to publish a notice of inquiry later this year seeking public input on further legal and policy topics, including how the law should apply to the use of copyrighted works in AI training and the ensuing treatment of outputs,” the guidelines states. Registration for these workshops is welcomed from artists, creative industry participants, AI developers, AI researchers, and attorneys. However, the Copyright Office has also mentioned in the published guidance today that there will be additional chances for stakeholders to weigh in outside of attending these meetings because attendance would unavoidably be limited.