Imagine That IP Law Blog

Artificial Intelligence (AI) is all the rage and grabs the public’s attention for its almost instantaneous song writing feats to creating deep fakes of all sorts. The possibility of having an almost instantaneous answer to a problem may prompt an employee to seek the answer AI can provide. Even if your company’s employee understands that the initial AI answer may be refined by follow up questions, this does not mean that the final result is not an intellectual property problem for your organization. Many companies believe their operations are such that the risk of IP infringement is low and there is no need for a formal AI policy. Some view instituting an AI policy as a response to a fad that is more likely to be an issue for education institutes. Others see a new AI policy as one more thing to for HR to maintain and police. I submit that it is best to institute an AI policy now and get ahead of the problem.

As noted, there has been a lot of publicity and multiple public conversations about AI and its surprising ability to rapidly answer questions on multiple topics. This leads many people to believe they understand AI and its ramifications, which in turn leads to a false sense of security with their use of AI. Many casual users do not understand the difference between generative AI models-which scrape the internet for multiple information sources to create a data base-and rules-based AI models-which generally do not scrape the internet and are typically company or application specific in terms of their data base. These distinctions between the types of AI and how they acquire their data base are of more concern to legal departments and, especially, intellectual property counsel or administrators then they are to the casual user.

Since rules-based AI models are less likely to be a problem absent some related issue like misappropriation of a trade secret, the focus here will be on generative AI models, such as ChatGPT, which offers a basic version for free. For the purposes of this discussion, we ran three different problems through an AI model: 1) drafting an HR policy on personal vehicle use and reimbursement for company employees; 2) drafting the layout for an advertising campaign for a new product; and, 3) preparing an instruction manual for use of the new product.

We start with the HR policy on personal vehicle use. The problem was outlined for AI in terms of when employee use is authorized, who can authorize the use, proof of insurance coverage, and a form for reporting the use and requesting reimbursement. In under a minute, this inquiry resulted in a two-page policy. See attached Personal Vehicle Use Policy) Admittedly, the result is somewhat bland and HR prefers more color and some graphics. Encouraged by the speed of the response, a request for photos or images follows, and some very generic looking photos are returned. A quick review satisfies the user that the results look very generic and are nothing special. Unfortunately, the selected photo may be from Getty images, Car and Driver, Road and Track, or the vehicle manufacturer’s brochure or advertisement. Likewise, the AI form may be from a legal or HR text and the selected information could be covered by copyright, such as a checklist for the identification of information for the form.   

For the advertising campaign, the problem was outlined as follows. Create an advertising campaign for new product that protects against sunburn, reduces skin irritation common aerosol suntan products, and offers longer lasting protection that know aerosol products.  Once again, the result (See attached SunGuard Campaign) appeared in less than a minute.  As a first concern, the suggested trademark, SunGuard, may have problems from an intellectual property perspective.  Next, there are medical claims regarding irritation and long-term damage, and possible false advertising issues regarding unmatched protection against sunburns, reduced skin irritation, and extended hours of enjoyment under the sun. Finally, the suggested use of visuals and videos again raises the possibility of getting material on line that is covered by third-party intellectual property rights.

For the instruction manual, the problem was outlined as follows. Prepare an instruction manual for a low temperature curing two-part epoxy with multiple hardeners. Here, the result appeared longer due to the vagueness of the request which required two clarifications; however, the final (See attached Epoxy Instruction Manual) took less than five minutes.  While the general nature of the AI generated manual suggest that it may have copied one of the many manuals published on line by the manufactures already in the market space, that is not the sole problem. The AI instruction to “Select the appropriate hardener based on the desired cure time and temperature range. Refer to the product packaging for guidance on hardener selection” may lead to patent infringement issues. Any one of the multiple manufactures in the relevant space may have patents covering the hardener selected, patents covering a specific combination of the two parts, and/or both. The patent infringement issues could include direct infringement and inducement to infringe based on the instructions.

In Andersen et al. v. Stability AI Ltd., 23-cv-00201-WHO (N.D. Cal. Oct. 30, 2023), a group of artists filed class action lawsuit against three AI-powered image generating tools (Stable Diffusion, Midjourney, and DreamUp), asserting the models powering these tools were trained using copyrighted images without consent.  The N.D. Cal. district court granted the defendant’s motion to dismiss all, but held that the claim of direct infringement against Stability AI could go forward.

In Thomas Reuters Enter. Ctr. GmbH v. Ross Intel. Inc., C. A. 20-613-LPS (D. Del. April 26, 2022), Reuters filed suit for copyright infringement alleging that Ross Intelligence used a third-party research company, which subsequently used Westlaw’s copyright protected headnotes, to train its natural language search engine. The district court denied the parties’ motions for summary judgment but did permit the filing of an amend complaint. Ross later alleged that it had closed the website, but it was apparently not confirmed and the case was ultimately order to trail.

In Doe 1 v. GitHub, Inc., 672 F. Supp. 3d 837, 847 (N.D. Cal. 2023), the plaintiffs sued OpenAI (Github) alleging Github wrongfully used their source code to develop their AI-based code-generation tools. Github’s AI tools (Copilot and Codex) were not programmed to treat copyright notices and license terms as legally essential, and reproduce licensed code used in training, violating countless open-source licenses. This case is instructive on the procedural issue as well as the intellectual property issues.

The above are a small sample of the current confusion and unrest in this emerging AI world.  As you might expect, what you get back for the AI model depends upon how you frame the question and where the model gathered the information. We have all seen the reports of the New York Times and other news organizations, as well as general interest and technical publications, raising copyright issues about their publications being scrubbed on line and used to train   generative AI models. We can expect to see more of this as organizations look to gather all of the revenue possible from their reporting and news gathering.

Many of the examples addressed above can seem trivial and would likely be caught by in-house counsel or possibly the head of human resources, marketing, or product development, if they were alerted to the issue.  However, despite the importance of the matter, counsel is often not made aware of an issue until after the fact. This can be especially missed in an organization where an individual may wear many hats. Having and publicizing an AI policy will, at the very least, alert employees to be mindful of the issues and seek the appropriate guidance when using AI for company business. The expense and business interruption that can accompany an infringement allegation or suit more than justify the effort to establish an AI policy now.

Reprinted with permission from the April 24, 2024 issue of The Legal Intelligencer ©2024 ALM Media Properties, LLC. Further duplication without permission is prohibited. All rights reserved.