Allison Aviki and Brian Nolan are featured in this article.
Users of generative artificial intelligence have seldom shied away from harnessing the technology to its fullest capabilities—fashioning everything from songs and paintings to poems and comic books.
Of course, once these creative outputs are generated, their human architects have naturally sought to protect them, using various aspects of the intellectual property infrastructure.
But they’ve all hit a snag. The law does not recognize AI to be a creator in the same way as a human being.
Indeed, the legal battles around AI and IP have gained significant traction in courts around the world. Just this week, the U.S. Supreme Court rejected a writ of certiorari petitioning an AI system to be named an inventor and qualify for a patent protection, thereby letting stand a lower court ruling that inventorship can only be awarded to human beings.
Still, so long as the technology is being used to generate works, the desire to safeguard them from open source use will likely remain strong. And to be sure, there may be loopholes in the legal framework to do just that. IP attorneys speaking at a “Protecting AI Inventions Through Intellectual Property” webinar hosted by Mayer Brown discussed the various legal avenues that currently exist, and the emerging viable pathways, to protect AI components and their IP.
Copyright claims are perhaps the more popular of AI protection strategies. Just last month, the U.S. Copyright Office launched its Artificial Intelligence Initiative, and is currently seeking guidance through listening sessions from experts.
If the monkey selfie and AI comic book cases are any indication, works generated by nonhumans cannot be copyrighted. However, the portion created by a human being is eligible for protection. Therefore, it’s all about knowing when an AI-generated work may have a decent shot of protection under the USCO.
Allison Aviki, a partner at Mayer Brown, said there are times that copyright protection is an AI creator’s best bet out of all other IP options.
“A familiar touchstone for consideration [of] copyright is when there are instances where the company is disseminating software externally, particularly with high probability or possibility for reverse engineering,” Aviki said. For example, by tracing the design process of the work in question, if an analysis can recover the human contributions to that work easily, it stands a better shot with the USCO.
Aviki noted that the USCO “has suggested” the prompts used to generate an AI-produced work might themselves be subject to copyright. However, this is dependent on the industry, the amount of expertise necessary to craft these prompts, and, of course, whether the prompts were substantially dependent on proprietary company data.
To be sure, at this stage, disclosure might be the most important aspect of applying to the USCO for copyright protection. Aviki pointed to the comic book case as example, where a registration was issued and then revoked in place of a partial registration once the USCO determined AI was used to create the work.
U.S. government officials, similar to those in the U.K. and Australia, have deemed that AI cannot be named an inventor. However, if a human being has sufficient involvement in the creation of the work in question, then the output may be considered eligible for a patent protection by naming the person an inventor.
For Brian Nolan, a partner at Mayer Brown, this means that companies must distill the contributions of “natural persons involved in AI to name as inventors” by following certain protocols.
In fact, it is very difficult to define a patent claim, even if it may just be a partial claim, if the entity is unsure about any portion of the AI’s workings.
“Companies should have policies in place that track the person better identifying the goals of the AI system, designing the AI system, generating the algorithm, compiling the data sets that train the model, testing the results of the AI output and implementing those results into products,” Nolan said.
Armed with this information, Nolan said companies may be better equipped to find the “nexus between a person’s contributions to AI components” and the claims presented in the patent application that may allow the company to identify the human beings involved as inventors.
However, this tracking process becomes more complicated, but even more important, as AI projects require input from various companies, individuals and entities. For example, one could provide an algorithm, while another company offers a specific field of data for the training set, and yet another contributes its own engineers.
Like the Coca-Cola recipe, protecting various components of an AI-generated output’s IP may be an option by going the trade secret route if the copyright and patent options are unviable.
However, Lori Recker, head of business operations at Aetion, said adequate measures must be taken for entities who may opt for the trade secret protection, and they aren’t always easy.
“Specifically, each component [of the AI work] can be protected provided that reasonable efforts are made to keep the information secret, and that the information derives independent economic value from not being generally known,” Recker said. “Assuming these things are true, then trade secrets may offer the best method of protection for components individually, because it allows protection in areas where patents or copyright may not.”
To be sure, one of the biggest risks to protecting an AI work using the trade secret protection is information disclosure. And this can happen intentionally or unintentionally.
“When trade secrets are exposed, it is most often through employees or ex-employees,” Recker noted. “So first and foremost, it is important to implement internal education and policies regarding the nature importance and treatment of IP.”
For the panel, contracts emerged as the most supple and flexible tool in the AI IP protection arsenal—just more limited in scope than its IP counterparts.
“Contracts can protect all components of AI, really,” Recker said. “But perhaps the downside is that the protection is limited to the parties that are subject to the contract and perhaps subsidiaries or affiliates or third parties that are included, but it is a more limited scope of protection in that regard.”
However, the strength in a contract’s pliability is evident in the various ways in which a creator or company may want the AI to be used or not used, questions of ownership and rights, and their timelines, Recker noted.
Like all IP protections, however, it is vital to keep track of the inputs and outputs that each entity involved in the contract might be contributing to the AI work, especially as questions of datasets around multiple outputs begin to emerge.
Reprinted with permission from the April 27, 2023 edition of Legaltech news © 2023 ALM Properties, Inc. All rights reserved. Further duplication without permission is prohibited.