When we think about AI and its “God-like powers,” we may think about how it may enslave us or destroy the world. Yet there’s a more subtle and prosaic connection between AI and the Divine, and, in fact, one that’s more likely to have implications right now – works created by either are ineligible for copyright protection under U.S. law.
You may have heard in the news that the New York Times is suing Microsoft and OpenAI over the use of their copyrighted material. In addition to the discussion of “fair use” of intellectual property, there is the question of how transformative human creative action can be. How do we relate the Divine and copyright?
On Wednesday, December 27, 2023, the New York Times filed a lawsuit against Microsoft and OpenAI. They argue that these companies, by using software to collect millions of articles from their website (a process called “scraping”) and creating products that can utilize their copyrighted material, are hurting their commercial relationship with their readers. Programs like ChatGPT can recreate entire written articles or summarize them, making it so that people do not need to subscribe to the New York Times, or even visit the paper’s website, to get the news it produces. Microsoft and OpenAI themselves did not even pay for access to their copyrighted work, they argued.
Additionally, The New York Times argues that their products were trained to treat their material differently, preferring or ranking New York Times material over other information sources. The algorithms were essentially told to prioritize information from the New York Times, indicating that the programmers felt it was more important/informative/valuable than other sources. If a particular content source is prioritized, it is more likely information from it will appear in text generated by ChatGPT in response to users’ questions. Because of this, the damage or harm done to the company’s brand reputation and reliability will also be increased when products like ChatGPT “hallucinate,” that is, provide incorrect or false information that can then be attributed to the “original source.”
This legal case will hang on how courts define the “fair use” of copyrighted material. Microsoft and OpenAI argue that including New York Times material in the training data sets for their AI is permissible because their use of it is “transformative.” The U.S. Copyright Office defines transformative uses as “those that add something new, with a further purpose or different character, and do not substitute for the original use of the work.” I hope it will be clear by the end of the discussion why I believe that generative AI fails to meet this definition.
There are four criteria for “fair use” that are typically used: purpose, nature, amount, and effect. Purpose is why the material is being used. Educational and non-profit purposes are typically seen as fair, while commercial or for-profit purposes are given more scrutiny. Nature is about what kind of copyrighted work is being used, with the distinction usually being made between creative and fact-based works. While the New York Times lawsuit concerns the latter, artists have lodged similar complaints about visual generative AI like DALL-E and Midjourney. Amount refers to how much copyrighted work is used, and the effect is how the use of copyrighted work affects the original work and its market.
I’m not a legal expert, so I won’t spend a lot of time analyzing all the details of fair use, but as a professor, the issue affects me when I quote or reference copyrighted material when I write, as well as in my syllabi and courses.
In the classroom, using an article or maybe a chapter of a book for free can be considered fair, but using large amounts of someone else’s work crosses the line. If a professor is going to assign several chapters from a book, they should ask their students to purchase it so that the copyright holder is fairly compensated. Compared to these educational uses, I would argue that Microsoft and OpenAI are doing something quite distinct from these academic practices, and based on my own reflection upon the meaning of fair use, they are indeed violating fair use.
Two theological reflections have arisen for me from this case. The first is that humans are creative beings, but how much of our creativity is truly original and not transforming something that already exists? Theologians have a variety of models that they use to define the image of God or highlight particular aspects of that image. One of the most common models of the image of God looked at the human capacity for reason and argued that this trait or ability is how humans reflected the image of God. While God possesses reason, God is also a creator. Lutheran theologian Phil Hefner coined the term “created co-creator” to describe his understanding of the image of God. According to Hefner, humans reflect the image of God in their ability to creatively act in the world. The word “created” taking priority over “co-creator” in Hefner’s definition serves as a reminder that humans are themselves the product of Divine creation and do not create out of nothing; rather, they transform God’s creation.
Philosophically, there are many ways people define ownership of something they’ve created. John Locke’s perspective was quite formative for how the United States defines property, and his Second Treatise on Government addresses the theological dimensions to this question. If God has given the capacity for creation to humans (which is something theologians debate about, but for the sake of simplicity, let’s assume this to be true), what is the relationship between creation and ownership? Locke argued that people are entitled to the fruit of their labor, and that this labor mixes with the natural materials with which we are working, resulting in a product that we can indeed own.
Locke’s caveat is that there has to be enough of the raw material left for others as well. This becomes even more complicated in a discussion about intellectual property and digital “material,” which lack any natural scarcity. The New York Times still “has” their material and owns the rights to it and its use, even after Microsoft and OpenAI scrape their data. Saving an image from the internet or taking a screenshot of something does not mean that you now have the only version or copy of it and thus “own” it, but in these cases the question about copyright is still about “ownership.” And for Locke, with an item that is not unique or scarce, the labor involved in its creation or transfer can help settle this question. If you buy a copy of Stephen Hawking’s book A Brief History of Time, you own that particular book, not ownership or the rights to reuse the content itself. If you google “Disney logo” and save the image of the logo to your computer, you do not own the image; that is not enough effort or labor to transform it into something that is yours. Does it really take effort for you to right-click and save a picture you like from a website? I do not think that would meet Locke’s definition of mixing your labor with a raw material. So while you might be able to “have” a copy of something, the rights of ownership still rest with the person who took the time and effort to create it.
In the case of how AI gleans this content for its reuse, it takes effort to create software that is capable of scraping, but I would argue that scraping alone takes even less human effort than saving a picture online, especially because once a program is created for scraping, it can be automated and used again and again on a variety of sources. In my view, transformation of the scraped content has to occur to invoke the fair use doctrine, and I am not convinced that what products like ChatGPT do with text is transformative. It would take longer than this piece to talk about how large language models (LLMs) like ChatGPT work, but as long as the AI is able to answer a prompt with the exact text from a copyrighted source verbatim, in my view no transformation of the material is involved.
The Divine and Copyright
U.S. copyright law affects religion and artificial intelligence quite similarly right now, as it places paramount importance on how something was created, and who did the creating. Many sacred texts are attributed directly to the Divine. The U.S. Copyright Office, in section 313.2 of its compendium, says that copyright protects works of human authorship, while works created by nature, including animals and plants, cannot receive copyright protection. The same is true for works whose authorship is attributed to a supernatural or Divine being.
Copyright protection can be given, however, if a work was said to be inspired by the Divine. So if a sacred text is said to be the direct words of the Divine, it would not be eligible to receive copyright protection. However, if it was said to be the result of Divine inspiration, then the work could be copyrighted. In the same section of the Compendium of Copyrightable Authorship (313.2), works created by machines without creative input from humans are also said to be not subject to copyright protection.
Matthew Allen had used the AI program Midjourney to create a piece of art that won the Colorado State Fair in 2022, but subsequently has been fighting a battle with the U.S. Copyright Office for over a year to establish his legal authorship. He argued that there was plenty of human creative action taken in the creation of the piece because it took “at least 624 text prompts and input revisions.” He had also used Adobe Photoshop and Gigapixel AI to edit elements of the piece, and to change its size and resolution. The U.S. Copyright Office said those aspects that had been modified could be protected, but not the entire piece as a whole. Allen plans to sue the government, but other legal cases surrounding AI and copyright protection have thus far been unsuccessful.
The current consensus on whether artificial intelligence can have authorship is that there has to be sufficient human input/effort in order for copyright protection to be extended to a given work. AI will continue to be at the forefront of copyright law, I believe, as the question of non-human creation and ownership will continue to challenge legal precedent.