“Artificial intelligence.” Many of you have heard this term. But what does it mean? Is it akin to “artificial sweetener?” In what sense, in other words, is AI artificial? Why should you care? And does it matter whether your brand is using AI in the E.U. as opposed to say New York City?

Some studies show that as many as 60 percent or more of executives in fashion are using generative AI. The use of this technology has grown in the fashion industry, like in some others. Now AI is being used to create designs along with content in a way not seen before. But there are different rules that apply for ownership of designs when using AI as opposed to, say, a Brooklyn based human fashion designer.
This article will provide an overview. The subjects are AI types, varied industry uses, and AI-related legal issues. It closes with some pointers to watch out for when deciding on whether or how much to use of this powerful technology.

1. Is AI artificial?
AI has been around since the 1950s. The term was originally birthed by Professor John McCarthy of Stanford. Since then, various types of AI have proliferated. The most common types now are generative and non-generative. These are important to consider for business. Non-generative or traditional AI generally involves supervised learning – where the system classifies a data point after seeing many labeled examples. Nothing new is created. This would be like a child being able to classify another sparrow after having seen many. Generative AI uses what is called unsupervised learning. This is akin to the child learning to classify a new bird even though he or she had not seen one before – after being trained on various types including sparrows, cardinals, and bluebirds. As such, new patterns can be created.

Notwithstanding the foregoing, some prominent computer scientists have critiqued AI. For instance, Professor Yoav Shoham, also of Stanford, has said, “AI has made truly amazing strides in the past decade.” However, “computers still can’t exhibit the common sense or the general intelligence of even a five-year-old.” Some studies from the likes of MIT have shown ChatGPT automation can increase efficiency. However, others at MIT point out the negative effects of using technologies like this one on originality. This in some cases results from the similarity of data sets for training. As a result, the type of AI training that you are using is relevant for copyright infringement, as discussed below.

2. Current uses
According to some sources, the use of AI is slated to skyrocket in fashion – to over 2 billion in 2032. This is relevant. The more that a fashion business likes yours leverages this powerful technology into the tech stack, the more there will be legal complexities. That’s because the technology changes very quickly. This isn’t academic. Generative AI is being used by some in the fashion industry to “create better-selling designs, reduce marketing costs, hyperpersonalize customer communications, and speed up processes.” The technology has also been used for trend forecasting. These uses give rise to the following legal considerations, among others.

3. Legal issues
A. Ownership
Ownership disputes can stem from using technologies like Chat GPT to create written content or similar technologies for designs. The E.U. and the U.S. generally follow the same line of reasoning when it comes to AI creations. In 2020, the E.U. Parliament passed a number of AI related resolutions. One goes on to clarify that “where AI is used only as a tool to assist an author in the process of creation, the current IP framework remains applicable.” See https://www.europarl.europa.eu/doceo/document/TA-9-2020-0277_EN.pdf. at pg. 7, ¶14 (emphasis added). Similarly, the U.S. Copyright office “will refuse to register a claim if it determines that a human being did not create the work.” (Emphasis added) The office further states that it will exclude works “produced by machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author.” (Emphasis added.) Consequently, AI-created designs are not likely subject to copyright registration.

B. Infringement
The foregoing implementations raise infringement liabilities. The first one is whether the inputs on which the model trains are being infringed. A second related one is the output: designs that are created with infringing inputs can be tainted, too.

With regard to the first, a recent case from the United States District Court for the District of Delaware is telling. The court found that a technology like Chat GPT which trains on data can nonetheless be infringing it – as such training is not a “fair use.” This means that, if a designer trains on other designs from the likes of Gucci, that training by itself can be infringing – even if the eventual design doesn’t directly incorporate any of these other designs.

As for the second, your company can be liable for the technology’s infringements—regardless of what the underlying technology is trained on. A recent example of this occurred when Google was held liable for its algorithms in their ability to potentially make connections and recommendations between users that enabled unlawful offline conduct. The Gonzalez v. Google case raises the issue of the “my AI did it defense.” While the case didn’t deal with infringement, the same reasoning can be applied to AI which creates infringing offline designs. A more recent case was brought against Singapore based Shein for using AI to scrape designs and make infringing copies.

C. Contractors
According to some studies, approximately 97 percent of manufacturing in the U.S. is done overseas. Within the U.S. many fashion companies outsource their works beyond their own company doors. As can be seen above, there are potential ownership and infringement liabilities in the subcontractor’s use of AI. Relatedly, there a slew of employment law concerns implicated, too, which are beyond the scope of this article. But they can affect legalities concerning ownership.

4. Closing
The AI market is moving with lightning speed. Like molasses, the law has been slow to respond. As a result, a recommended practice is to have a clear understanding of the technology and applicable legal uncertainties via your chief technology officer and general counsel, respectively, before fully implementing.

Ryan is the principal attorney at Long & Associates PLLC, a Malibu based boutique intellectual property (“IP”) practice. Since 2006, he has navigated pioneering brands like Apiece Apart and CFDA members through tricky IP related legal disputes —including trademark ownership and manufacturing liability. As a non-residential fellow of Stanford Law School’s Center for Internet and Society, he has written for, or has been interviewed by, the likes of the TechCrunch about AI legalities. His e-mail is: rlong@landapllc.com.