Welcome back to the Data Connections Blog!
Did you realize that there is an existential legal battle being fought right now over data? The outcome of this battle may determine the future of both Artificial Intelligence and the creative arts. With all the other sound and fury in the current ranks of power, it might be easy to miss the importance of this point in history.
Am I overstating this for dramatic effect? I don’t think so…. but, please read on and let me know what you think.
Unlike prior administrations, The Trump administration’s actions to date indicate more interest in the development of powerful US AI systems than in concerns for the need to create guardrails for those systems. Out of the gate, the Trump Administration rejected the Biden-era AI safeguards.[1] This coincided with private and public meetings between President Trump and the leaders of major AI creators, like Sam Altman of OpenAI.[2]
One of the consistent themes in the discussions with these AI industry leaders is the need to allow for more and more protected training of AI systems on copyrighted data.[3] If you recall the movie or play Little Shop of Horrors, AI is a bit like the Audrey II character. Audrey II is a plant that must constantly be fed human flesh and blood. An apt analogy for AI, since its creators are looking to feed it with the blood, sweat and tears of artists and creators who have come up with the books, movies, articles, etc., that are the food for AI systems.
As a refresher, the original ending for the Little Shop of Horrors movie had Audrey II and its progeny growing exponentially out of control and eating everyone, ultimately vanquishing all humankind. It was not well-received by audiences in screen tests and so the ending was changed to one in which Audrey II is vanquished by humans instead.[4]
Putting aside the vanquishing for now, let’s look at the exponential growth part of the analogy. In my Data Law class, I tell the story of providing legal support to a data group that ultimately would create a major commercial AI system. This was about five years before it became a well-known commercial system. After months and months of swamping my legal team with constant and frequently last minute requests to negotiate agreements for datasets (typically quite one-sided in favor of the creator) and to “clear” the data for ingestion into our company, I scheduled a meeting with the leader of the team to see if we could come up with a better approach.
I asked the head of the data team to help me understand their priorities for datasets so we could try to balance out the workload all while still meeting their requirements. She calmly responded “…all of it; we want, and we need, ALL the data we can get.” Okay… not that helpful but at least I understood. So, I thought maybe we could try to work with the timing. I responded, “fair enough, but can you help me to understand your priority and timing for when you must have different types of data.” Again, the response was instant if not terribly helpful for my purposes: “yesterday, we needed it all yesterday, but we can probably live with getting it tomorrow.”
At some level I understand why AI leaders, and AI creators in general are pushing to knock down any perceived walls between data and their AI systems, even if I don’t necessarily agree that all copyrighted materials should be available to them instantly and for free.
The argument that the AI creators are making is that their usage of any data covered by copyright is permissible under the fair use provision of the US Copyright Act. Since virtually every person who has ever wanted to use a copyrighted material without payment has argued that their proposed use is a “fair use” it probably makes sense to dig into just what that term means.
To start, the fair use provision of the Copyright Act allows you to copy and use copyrighted material for purposes such as: 1) criticism, 2) comment, 3) news reporting, 4) teaching, 5) scholarship, or 6) research.
Is it really as easy as just saying that when we are training an AI model, we are teaching that model… or can we just name our AI company the HAL Research Institute and get “all you can eat” use of data under the fair use exception to copyright?
Well, no. It is not quite that simple (and it should not be). Simply slapping the “research” label on an otherwise fully commercial usage is not likely to make it a fair use. The determination of whether a use is fair is undertaken using a four factor test which considers the following...
- 4 Factor Test:
- 1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
- 2) the nature of the copyrighted work;
- 3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
- 4) the effect of the use upon the potential market for or value of the copyrighted work.[5]
- 1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
In practice, a fair use defense against copyright infringement is highly fact dependent and it is difficult to predict the outcome. High profile cases go through many levels of appeal and remand for reconsideration, and settlement is possible at any stage.
Most recent fair use analysis by courts has focused on factor 1 above, which considers whether the proposed usage of the copyrighted material could be considered transformative.
As described by the US Copyright Office:
"Transformative uses are those that add something new, with a further purpose or different character, and do not substitute for the original use of the work."[6]
Let’s take a look at two examples. In 2015, when Google digitized a large volume of copyrighted books using that data to enable “snippet” searches of the books (not showing the full view), its usage was determined by the Second Circuit to be transformative and a fair use.[7] Essentially, the finding was that snippet searching was a new purpose, and that a snippet was not a commercial substitute for the original work (Hmm, using Google snippet searches I should be able to read the entirety of the new Sunrise on the Reaping novel for free in about 3 years… think of the money I will save!).
By contrast, in 2023 the U.S. Supreme Court found that an Andy Warhol painting based on a Lynn Goldsmith photograph of Prince was not a transformative fair use. Despite extensive similarity in the two images, the Warhol estate argued that the Goldsmith photo showed the “Vulnerable Prince” while the Warhol version transformed him to the “Icon Prince.” The Court found details in the record of the case showing that the photograph and the painting served a similar market and that both had actually been considered for the same magazine cover. Because the Warhol work shared the same commercial purpose as the original photograph, the Court found that that this was not a fair use. [8]
With this background (and… yes there is a lot more background on fair use that I cannot cover in this space!), do AI creators need to worry about using data to train their systems? The clear answer is, as all good lawyers know, “it depends”.
Training AI on massive amounts of data (either discrete sources or scraped[9] from the web) enables the AI to learn from the data’s “… patterns, context, and structures, creating transformative outputs….”[10] But just exactly how transformative are those outputs?
In various talks I have given on data, I use the example of a farmer in Southern Ohio who asks an AI system to recommend when it would be most optimal for her to plant soybeans. The AI system may use highly factual information including NOAA weather data, historic crop yield information, Chicago Mercantile Exchange crop pricing data and the like to yield an answer like “anytime on or soon after April 15th”. That is likely a transformative use of data since the answer in no way impacts the commercial value of the underlying data and probably doesn’t even contain any of the original data in it.
At the other end of the spectrum, we know that AI systems are being trained on large volumes of very original copyrighted materials. So much so that when I asked ChatGPT:
Create a children’s Halloween pop-up book in the style of Stephen King
...in a matter of seconds it does so:
Is this transformative? How much of Stephen King’s original expression made it into that story? It’s hard to tell but it may be more than we think. That said, let’s consider this from another perspective. Copyright exists to promote the progress of science and the useful arts.[11] The fair use exception also promotes the progress of science and the useful arts by encouraging commentary, research, and new transformative uses of existing materials. In the age of AI, do we want to allow fair use to consume the original purpose of copyright when it comes to content creators?
Do we even need authors or artists when we already have enough of their existing works ingested into AI systems that, in a matter of minutes, we can get a book generated in the style of a particular author but in a relatively new format (Stephen King has only written one, fairly recent pop-up book that I am aware of[12] )? This new work might easily displace the ability of that author to enter further into this market. I mean who needs to pay for a Stephen King pop-up book when I already have one for free?
More telling is that when I asked ChatGPT to create a short story in the style of Stephen King, a genre where he has produced many works, I was once again able to get a story in a matter of seconds.
For content creators and artists of any type, the concerns over AI are far more significant than my simple examples might indicate. In comments submitted to the US Copyright Office regarding the need for reform of fair use in light of AI developments, the Screen Actors Guild – American Federation of Television and Radio Artists (SAG-AFTRA) noted "the artistry and skills of humans as paramount to creativity and creation" [emphasis added]. SAG-AFTRA further emphasized that "we are rapidly barreling toward a future where creators will have to compete with unauthorized versions of themselves or their works."[13] (Hmm, sounds like an existential crisis to me!)
There are numerous lawsuits currently in the courts pitting creators versus AI systems. The table below only gives a small sampling of these, but the themes are consistent. Artists claim whole cloth use of their copyrighted materials (be it films, music, books, news articles, etc.) and the AI creators either deny such usage and/or argue that their usage was fair use.
| |||
|
|
|
|
| Andersen v. Stability AI (ND CA 2023) | Copyright Infringement (CI), Right of publicity, DMCA | Only direct infringement claim survives; court says AI is legally distinct from prior tech |
| Dow Jones v. Perplexity (SD NY 2024) | CI via Retrieval Augmented Generation (RAG) tech (AI copies data from other AI models) | Alleging repackaged and verbatim content of Dow Jones in AI; focuses on RAG |
| Kadrey v. Meta (ND CA 2023) (consolidated with Chabon, Farns-worth vs Meta) | Infringement via copying of Books3 dataset | Meta admits Books3 use in training but argues fair use |
| UMG v. Suno / UMG v. Udio (D Mass 2024) | CI of recordings (pre/post-1972) | Defendants admit copying; claim fair use |
Time will tell how these cases are resolved. Like the Little Shop of Horrors movie, there are at least two potential endings to this copyright battle. The first ending, akin to the one ultimately produced in Little Shop of Horrors, sees creators and artists taking back control over the AI threat to their works, metaphorically killing the monster and living happily ever after. In this scenario, the courts and legislatures help to ensure that AI developers obtain proper authorization and pay compensation for the use of copyrighted materials.
The second ending is reminiscent of the original ending in Little Shop of Horrors (the one that wasn't used). In this scenario AI, much like Audrey II, grows and reproduces and ultimately takes over the [creative] world. This would mean that AI technology continues to advance rapidly, and fair use is interpreted broadly, allowing AI developers to use copyrighted materials with minimal restrictions. This outcome could lead to an explosion of AI-generated content at a potentially significant cost to human creators… completely reshaping the creative landscape.
Finding a balanced path forward requires collaboration and compromise between AI developers and content creators. Establishing clear guidelines and standards for fair use in the context of AI is essential to protect the rights of content creators while fostering innovation in AI technology. Additionally, exploring new models for compensation, such as licensing agreements or revenue-sharing arrangements, could help bridge the gap between the two parties.
Ultimately, the resolution of this existential copyright battle will shape the future of the creative industries and the development of AI technology as both sides navigate this complex issue.
I hope you’ll be back next time as we continue to explore these issues and learn more about data connections!
If you have questions about your data and your legal compliance programs for data, Mortinger & Mortinger LLC can help! Contact me directly at: steve@mortingerlaw.com.
https://www.cnn.com/2025/02/28/tech/trump-ai-apple-investment-stargate-deepseek/index.html ↑
- https://www.forbes.com/sites/virginieberger/2025/03/15/the-ai-copyright-battle-why-openai-and-google-are-pushing-for-fair-use/ ↑
- A copyright protects original works of authorship (like books, paintings, photos, drawings, music, computer programs, movies, plays, etc.) created by a person in a fixed / tangible form of expression. It is not a protection for pure facts. ↑
- https://www.imdb.com/title/tt0091419/alternateversions/ ↑
https://copyright.columbia.edu/basics/fair-use.html ↑
https://www.copyright.gov/fairuse/#:~:text=Additionally%2C%20%E2%80%9Ctransformative%E2%80%9D%20uses%20are,original%20use%20of%20the%20work. ↑
Authors Guild, Inc. v. Google, Inc. (2d Cir. 2015)
- Andy Warhol Foundation vs. Goldsmith (S. Ct. 2023) ↑
- Web scraping involves extracting data from web pages, typically using software bots and pulling that data into an offline location where it can be used. This is not a new concept and dates back to 1989. ↑
- https://www.forbes.com/sites/virginieberger/2025/03/15/the-ai-copyright-battle-why-openai-and-google-are-pushing-for-fair-use/ ↑
- US Constitution Article 1, Section 8, Clause 8: https://constitution.congress.gov/browse/essay/artI-S8-C8-3-1/ALDE_00013063/ ↑
- The Girl Who Loved Tom Gordon: https://www.stephenkingcatalog.com/catalog/girl-who-loved-tom-gordon-popup-p-123.html
- https://copyrightalliance.org/generative-ai-ethical-training-practices/#:~:text=Unchecked%2C%20we%20are%20rapidly%20barreling,future%20human%20creativity%20and%20expression.%E2%80%9D ↑
https://copyrightalliance.org/ai-lawsuit-developments-2024-review/#:~:text=The%20case%20involves%20a%20group,Product%2C%20DreamStudio%2C%20and%20DreamUp. ↑