Texas Bar Journal • May 2024

Generating Content

Can artificial intelligence create IP for itself and for you or your clients?

Written by Marc Vockell

As an intellectual property lawyer focused on artificial intelligence, I learn about impactful—and sometimes surprising—developments nearly daily. Interesting news in the law, business, and technology of AI pops up almost as fast as a ChatGPT answer to a prompt for a poem (if you haven’t tried, it’s fast). Therefore, this article will not attempt to provide the current update because it would be outdated by publication. It instead offers a framework for IP analysis that will endure as the technology, business, and law develop.

To stay current on the dizzying updates on AI and IP, the internet is the obvious resource.¹ For reliable updates on what’s happening on the cutting edge of IP and AI, the U.S. Copyright Office and U.S. Patent and Trademark Office, or USPTO, each provide updates on their websites, and they send free email updates to anyone who subscribes.² They have also been accepting and posting public comments, providing town halls, and offering other resources. Podcasts also provide technical and business updates on AI. For example, Hard Fork covers technical and business issues, while TWIML AI Podcast is more technical, and The AI in Business Podcast focuses more on business.³

Artificial intelligence has been around for decades and has focused on applying algorithms to exponentially growing datasets to try to mimic, or at least assist, human decision-making, such as by identifying or predicting the presence of birds, tumors, or other phenomena in various contexts. Through improvements in machine learning, the AI tools now can utilize data and algorithms to “teach” themselves how to better achieve goals.

AI leaped forward in late 2022 when generative AI, or GAI, flew into headlines through the emergence of ChatGPT, or generative pretrained transformer, and similar platforms. GAI applies trained algorithms to vast datasets to create new content, employing neural networks. GAI answers prompts provided by users in a process that repeatedly predicts what each “next word” is in response to a text query. This makes “autocomplete on steroids” a helpful analogy for GAI’s current state.⁴ GAI doesn’t just predict the immediate word, phrase, or search you are typing; it serially (and astonishingly quickly) predicts each consecutive word as it creates an entire memorandum, letter, essay, poem, song, or other creative work. GAI is not just for text; for example, it can also predict what each next pixel should look like to create an image, or what the next note should sound like in a musical composition. GAI has been successful at quickly creating content that is creative, if not accurate—yet. Given its impressive ability to rapidly create, it raises interesting intellectual property issues.

CAN AI CREATE PROTECTABLE IP?
A threshold question is whether GAI can be a creator or inventor of IP. The early test cases have yielded rulings by the Copyright Office and the U.S. District Court for the District of Columbia that output of generative AI is not copyrightable because there is no human authorship.⁵ Even if a human claims to be the author by entering the query into the AI tool, AI output is not copyrightable.

This is true even if the human employing the AI tool edits the output, as the human author did with the visual output of the graphic novel Zarya of the Dawn.⁶ It is also true if the human goes through hundreds of iterations of inputs to secure the output, as was done with Théâtre D’opéra Spatial, a piece of art that won first place in the Colorado State Fair but was denied registration by the Copyright Office.⁷ Similarly, outputs of AI are not copyrightable if the human inputs an image and asks for it to be transformed into a form such as Van Gogh’s Starry Night.⁸ Compilations of work that include elements created by AI may be copyrighted even if individual images or portions of text are not.⁹ And works that disclaim portions created by AI can be copyrighted.¹⁰ It is also possible that detailed input prompts can be copyrightable.

Considering other types of intellectual property, GAI cannot be an inventor on a patent. The U.S. Court of Appeals for the Federal Circuit upheld the USPTO’s determination that a patent could not be issued to an AI tool because an “inventor” must be a natural person, and the U.S. Supreme Court denied certiorari.¹¹ It is still an open question as to what extent and how a human inventor may utilize AI to assist with the inventive process. The recent executive order on AI directed the USPTO to provide guidance on the topic, which was issued February 2024.¹²

Trademarks are a different story. There is nothing prohibiting trademarking a slogan or logo created by AI. But as with any trademark, the issue is whether it is a distinctive identifier of goods or services.¹³ Utilizing a logo or slogan created by GAI is risky because there is no way of telling if competitors might apply similar prompts that could lead to confusingly similar trademarks.

Trade secrets have a more complex and fact-specific relationship to GAI, both in relation to input and output. Inputting trade secret information into a GAI tool likely waives the trade secret, possibly even if confidentiality provisions purportedly apply. Receiving useful output from a GAI tool may arguably provide information that is protectable as a trade secret because trade secrets don’t need to be created by humans.¹⁴ But this could only work if the terms, conditions, and processes of the GAI tool provide not only ownership (discussed below), but also security and confidentiality in providing output.¹⁵ Even then, courts or juries may disagree.

BETWEEN THE USER AND AI TOOL, WHO OWNS WHAT? IMPORTANCE OF TERMS OF USE AND PRIVACY POLICIES.
It is important to remember that as highly technical and cutting edge as AI might be, its use is still essentially transactional. A user makes an agreement with an AI tool to provide an output. It is therefore essential to study the terms and conditions of any AI tool.

Each tool has its own varying terms.¹⁶ The free version of OpenAI’s ChatGPT, for example, defines its “content” as consisting of the “input” provided by the user and the “output” provided by ChatGPT.¹⁷ OpenAI further states that the user not only owns the input, but also (subject to compliance with terms) assigns OpenAI’s rights, title, and interest in the output to the user. It further states users may use the output for any purposes, including commercial purposes. The terms say OpenAI may use the content to “provide and maintain” its services, which would appear to give OpenAI the right to use the input and output for training purposes. OpenAI disclaims liability and offers no indemnity for use of output it creates for “free” users. On the contrary, OpenAI’s terms say users indemnify and defend OpenAI against any claims “arising from or relating to your use of the Services.”

The GAI art tool Midjourney, for another example, provides differing IP rights to output depending on whether usage is free, subscription, or corporate.¹⁸ The free subscription provides a user with a “Creative Commons NonCommercial” license to output, while the paid subscription provides ownership of output to the users. Use on behalf of companies with more than $1 million annual gross revenue requires a “Pro” or “Mega” membership.

Regardless of user type, Midjourney states that users provide a broad, perpetual license to Midjourney in any text and image prompts a user inputs.

If an AI user wants to utilize outputs of GAI for commercial purposes and minimize risk of liability for infringement, indemnity clauses are important. Use of outputs of GAI can create risks not only of copyright infringement, but also for violations of rights to publicity or privacy. Some GAI providers such as Google, Adobe, and others offer certain types of indemnity for services— usually for paid subscription services.¹⁹ OpenAI announced in November 2023 that its “Copyright Shield” will defend and indemnity its developer and enterprise ChatGPT users.

It is important, however, to look carefully at the details of indemnities because there are a variety of limitations, conditions, and exceptions. For example, Microsoft provides indemnity for commercial Copilot users and “will defend the customer and pay the amount of any adverse judgments or settlements that result from the lawsuit, as long as the customer used the guardrails and content filters we have built into our products.”²⁰ Adobe’s enterprise indemnity “will not cover claims to the extent that any claim is based on or arises from the Customer’s action or content,” listing five bullets as examples “without limitation.”²¹

CONCLUSION
The technical, business, and legal aspects of GAI will continue to evolve rapidly. Continued review of the latest terms of use of GAI providers is imperative. Awareness of issues relating to who owns the outputs of AI from an IP and contractual basis will help us mere human users maximize IP value and minimize risk.

MARC VOCKELL is responsible for intellectual property including trademarks, copyrights, patents, sponsored research, clinical trials, technology commercialization, software licensing, confidentiality, and trade secrets at the University of Texas System. He previously worked at Dell Technologies, where he served as vice president for intellectual property litigation. Before working at Dell, Vockell was a business litigation attorney at Vinson & Elkins. After graduating from Purdue University, he taught bilingual elementary education in the Houston Independent School District, through Teach for America. Vockell then got his law degree from the University of Texas School of Law in 1997.