Understanding US Copyright Policies in the AI Era

The intersection of artificial intelligence (AI) and creativity has opened up a new frontier in the world of marketing, with companies increasingly exploring the use of AI-generated art in their campaigns. However, the recent ruling in the case of Théâtre D’opéra Spatial by the U.S. Copyright Office (USCO) has introduced a new layer of complexity and uncertainty. The decision reaffirmed that AI-generated works are ineligible for copyright protection (at least for now). This POV explores the implications of this landmark decision for companies considering using AI-generated art in their marketing strategies.

Let’s delve into the intricacies of the ruling, its grounding in the principle of human authorship, and the broader implications for various forms of creative output beyond visual art.

By examining the potential risks that companies may face, from the lack of copyright protection for AI-generated art to potential infringement of other rights, if the AI has been trained on copyrighted works, we can explore the alternative forms of intellectual property protection that companies might consider and the ethical and public perception considerations that come into play when using AI-generated art.

In an era where AI is becoming an integral part of creative processes, we aim to clarify the evolving landscape and offer guidance for companies at the forefront of this technological revolution.

SUMMARY OF THE DECISION

The USCO Review Board has reaffirmed its stance that works generated by artificial intelligence cannot be copyrighted without sufficient work or authorship by humans. The decision was made in response to Jason M. Allen’s second request for reconsideration of the Review Board’s refusal to register a two dimensional artwork titled Théâtre D’opéra Spatial, which he created in part using an AI system. The Review Board maintained that the work contained more than a de minimis amount of AI-generated content, which should have been disclaimed in Mr. Allen’s application for registration. As Mr. Allen was unwilling to disclaim the AI-generated material, the Review Board upheld the denial of registration.

The Review Board’s decision is rooted in the principle that human authorship and creation is the bedrock of copyright protection. The decision leaves open the possibility for further analysis by the USCO on a case-by-case basis, and there are expected to be further developments in this area of law.

WHAT PARTS OF THE WORK WERE ELIGIBLE FOR COPYRIGHT?

According to the USCO’s decision, while the work as a whole could not be copyrighted based on the facts of this case, it did try to clarify that certain human-authored modifications to the AI-generated material may potentially be protected by copyright. This would be the circumstance if the human author’s contributions involve sufficient creative choices and originality.

*Does this ruling apply only to images or more broadly? *

The USCO’s decision specifically addressed a case involving a two-dimensional artwork. However, the principles outlined in the decision—particularly the requirement of human authorship for copyright protection— may apply more broadly to all forms of creative expression, including written works. They could also help inform how businesses wishing to protect copyright should proceed.

The USCO has stated that “the Office will refuse to register a [copyright] claim if it determines that a human being did not create the work.” This stance will likely extend to written works generated by AI.

However, as with visual art, human-authored modifications to AI-generated text may be copyrightable if they involve sufficient creative choices and originality. The specifics of each case would be significant, and the USCO would need to consider whether the human-authored elements can sustain a claim for copyright.

WHAT OTHER FORMS OF PROTECTION ARE AVAILABLE TO BRANDS?

While the great majority of AI-generated works may not currently be eligible for copyright protection in the U.S., there are several other forms of intellectual property (IP) protection that companies should consider (with the help of counsel):

Trademarks: In some cases, artwork used in conjunction with the brand could be eligible for trademark protection. This would offer protection against other entities using a similar mark in a way that could confuse consumers. Additionally, any logos, brand names, and other elements of marketing campaigns that are already trademarks would continue to be protected—meaning that while some AI-generated elements of a marketing campaign may not be protected, the overall advertising units (consisting of art, logos, and copy together) are protected in the configuration used.

Trade Secrets: The prompts, algorithms, data sets, and techniques used to create AI-generated art could potentially be protected as trade secrets. This would require the company to take steps to ensure the secrecy of these elements, such as implementing strict access controls and confidentiality agreements. Keeping prompts as trade secrets makes it impractical for the AI-generated artwork to be reused by others.

Patents: Companies may be able to patent the unique methodsor systems used to generate AI-created art. However, patent law also requires human inventorship, and obtaining a patent can be long and complex, and generally requires the assistance of specialized legal counsel.

HOW CAN MARKETERS USE AI-GENERATED CONTENT?

*Does a lack of copyright mean that AI-generated content is useless for AI marketing? *

Not at all. AI-generated content can be used as long as risks are properly managed.

Firstly, the brand has the right to use the AI-generated work explicitly because it is not copyrighted. No license fee or royalties need to be paid. The risk is that anyone else could use the AI-generated work you create too.

Generally speaking, all ads will have the brand name and logo placed “on top” of them at some location. The brand name and logo are likely trademarks and protected separately from copyright. No one else has the right to use the ad in its entirety other than the trademark owner.Brands will need to weigh the residual risk of the AI-generated content against its utility to decide about the elements “underneath” the logo and brand name.

There is some nuance around which AI uses matter. It helps to consider this using examples.

EXAMPLE 1: GENERATIVE-AI FILL

Consider a brand that ran a photoshoot in a desert. The photos from that shoot are copyrighted, and the copyrights are held by the brand. The photo, as seen in Figure 1, has too many footprints. An artist uses an AI “generative fill” algorithm to smooth out the sand and limit the number of footprints.

Figure 1 - Original Photo and AI-Fill Results (courtesy of Adobe)

What is the utility? A cleaner image was made with minimal effort.

What is the risk of reuse? The AI-generated sand and footprints cannot be copyrighted. In principle, the footprints could be copied and reused by another party. In practice, it would be virtually impossible to determine which parts of the picture are reusable without access to the details held within the AI tool. Even if these could be disentangled, is there any actual damage done to the brand by the reproduction of the footprints?

Likely not. In this case, this use may be safe for the brand under specific circumstance.

EXAMPLE 2: GENERATIVE-AI EXPANSION

In this example, the brand conducted a photoshoot in the mountains (Figure 2). The brand owns the copyright of the photos. After the shoot, there was a desire for a wider angle shot than was captured.

An artist can use generative AI to expand the image. In this use, the AI fills in the area beyond the edge of the image using a probabilistically likely scene. It blends perfectly with the actual photo.

Figure 2 - Generative-AI Expansion (courtesy of Adobe)

What is the utility? A wide-angle version of the image was made available with minimal effort. Even reshooting on location would not achieve this, as changes in lighting, seasons, foliage, etc., would never match perfectly.

What is the risk of reuse? The AI-generated edges of the image cannot be copyrighted. They could be re-used by someone else. The core photo is still copyrighted, and trademark protection likely applies to any logo and brand name used. This comes down to a risk judgment as to the value of the edges of the image.

In this case, we would argue that the use of AI-generated content is of very low risk to the brand and is very likely acceptable.

EXAMPLE 3: GENERATIVE-AI FOUNDATION

Figure 3 shows the work at the heart of this copyright decision. The AI called “Midjourney” was used to generate the foundational part of the art on the left. A human artist then modified it to create the final work on the right.

As we have seen, only the human-applied edits can be copyrighted. The underlying AI-generated image cannot be copyrighted.

Figure 3 - Théâtre D’opéra Spatial

What is the utility? A beautiful image was created and customized to the needs of the brand.

What is the risk of reuse? Here, there is certainly some risk of reuse. In this instance, the AI-generated content is fundamental to the overall piece. Compare it to the use of stock photography with a non-exclusive license. The same background image might appear in other places and be used by other companies.

The AI-generated content is effectively a zero-cost stock photo, and for campaigns not deeply tied to the underlying image, this may be an acceptable compromise for the brand. Conversely, if the image is the “hero” of the brand meant to stick in customers’ minds, this use is probably unacceptable. It is an instance where judgment matters.

RECOMMENDATIONS

At Klick Health, we believe that AI-generated content is too big of an opportunity for brands to pass up. We recommend the creation and communication of a policy defining when AI-generated content is acceptable and when it is not.

Specifically, we recommend:

counsel to register and protect their logos and brand names as trademarks

their legal teams to create a policy on AI-generated content

agencies and any other content authors working on behalf of the company

currently associated with the use of AI fill and AI expansion and consider the practical impact

This policy can and should be extended for uses other than visual art, such as the written word.

ADDENDUM NOVEMBER 1, 2023: AI TRAINING COPYRIGHT LAWSUITS

There is a second, unrelated set of copyright litigation underway around image-generating AIs. These cases dispute whether or not specific AI platforms violated the copyrights of authors and artists when creating their systems. This does not have to do with the use of the images generated by the system but with how the system itself was created. Multiple cases are proceeding in the U.S. and other countries. A recent ruling in the U.S. set a high bar for these plaintiffs to meet.

We welcome your questions and feedback. Please contact: Michael Chambers SVP, Opportunity Creation at mchambers@klick.com