UNM Rainforest Innovations

Below is a piece written by Charles Valauskas of Valauskas Corder LLC, who serves as patent counsel for UNM Rainforest Innovations. In this piece, he touches on the potential risks and limitations associated with artificial intelligence large language models, including ChatGPT, especially regarding intellectual property protection.

Generative artificial intelligence (AI) large language models are innovative tools. The most popular of the large language model-based chatbots – Chat Generative Pre-Trained Transformer, or ChatGPT – was developed by OpenAI and launched on November 30, 2022. Its almost immediate popularity encouraged other companies to introduce their own versions including Bard and Ernie Bot. Each of these tools can provide interesting narratives, or write lyrics and poetry, compose music, write and help debug computer programs, answer questions including those offered as part of an exam, and play games.

However, is a chatbot the right tool for all applications? Clearly not. Using one can have many serious unintended consequences. The following will focus on the use of chatbots and what may happen to the intellectual properties that may be associated with the information that a user submits to a chatbot to start the human/machine “dialogue.”

First, recognize that the creative thought or expression that a submitter wishes to take on a test drive on the Chat expressway may have some value, possibly simply because who submitted it, and the full extent of which may not be fully known at the time of submission. That thought or expression, if not generally known by the public, may qualify as the most basic and important of all intellectual properties: a trade secret. Many wish to call this information, having value that is not generally known to the public, “know how.” However, regardless the label attached to it, the information can be protectable under state and federal trade secret law.

To maintain the protection of information under trade secret law, the information must not be released to the public. However, even though some generative AI tools often claim to have some privacy safeguards, submitting information to a chatbot may constitute a “publication,” that is, a release to the public of the information. As a result of, and automatically upon the submission to the machine side of the dialogue, the ability to characterize and hold the submitted information as a trade secret is lost. 

Another unintended consequence follows from the publication of the submitted information. If the submitted information described an invention, in part or in whole, the publication can automatically eliminate the possibility of obtaining a patent in nearly all foreign jurisdictions. For example, European patent law provides that a European patent application must be filed prior to any publication of the details of an invention in order to preserve the opportunity to obtain a European patent for the invention. The publication also causes a clock to start in the U.S. A patent application for the invention must be filed with the U.S. Patent and Trademark Office no later than a year from the date of publication to preserve the opportunity to have a U.S. patent issue for the invention.

The above shows that the simple submission of information to a chatbot to start a conversation with it about the information may cause the unintended loss of two important intellectual properties, trade secret and patent.

The submission may prevent the creation of another important intellectual property. Users expect that, by simply submitting a question, the chatbot will provide a narrative. The machine-generated narrative may be interesting yet can the person who submitted the question to the machine claim that the resulting machine-produced narrative to be the person’s work protected under the U.S. Copyright Act (and under Treaty law in major foreign jurisdictions). The answer is possibly not. The Copyright Act makes clear that “copyright protection subsists… in original works of authorship fixed in any tangible medium of expression….” The key word in this statutory provision is “original.” Courts have defined “original” as meaning not copied from someone else. Clearly, because artificial intelligence systems train on using other people’s content to develop the narratives that submitters request by their submissions, those narratives by definition are copied in part or in whole from someone else. As a result, there is a good chance that the narratives produced by AI systems are not owned by the person submitting the question. Building on a narrative, such as software code, or music, or lyrics, or a screenplay whose ownership is not known, would present a challenge that is difficult to overcome.

Users who choose to use publicly available generative AI tools should understand the potential risks and limitations associated with these versions. At the very least, confidential information or trade secrets should not be submitted to an AI system given that the information will likely be used and published in some form and without the submitter’s knowledge. Remember that once data is entered into a generative AI tool, it cannot be removed.

Learn more about intellectual property protection and details on disclosing an invention here: https://innovations.unm.edu/protecting-intellectual-property/

Skip to content