You Won’t Let Those Robots Eat Me


“How do you thank the idea that history is alive? This book simply would not exist if I had not read other books and encountered ideas in a kind of constant flux, a little like swimming in the ocean…books just don’t materialize from an author’s head, nor exclusively from research and creative labor. The making of books is a creative collaboration, and this book owes it’s very life to several radiant souls whose lives threaded through mine.”

Thrust, by Lidia Yuknavitch

On January 13th, artists Karla Ortiz, Kelly Mckernan, and Sarah Andersen filed a lawsuit in federal court against artificial intelligence (AI) companies Midjourney, Stability AI, and DeviantArt, claiming that they violated copyright laws by using the artists’ images, along with images made by tens of thousands of other artists, to produce derivative works used to train the defendants’ AI products. In other words, the artists say the works generated by the defendants’ AI infringes their copyrights because the ultimate works are mere derivatives of the training images, several of which are works created by these plaintiffs. The plaintiffs allege that any artist whose images were used by these companies as training images are also part of the class of injured persons.  

This past year saw an explosion in AI tools used primarily by artists and creatives, but increasingly also by amateur artists and people to create images by sending written prompts to an AI tool such as Midjourney. 

To understand, consider this example. David Hockney is an important contributor to the pop art movement of the 1960s, and considered one of the most influential British artists of the 20th century. I also love his paintings!  Here are a few images that you might recognize from Hockney: 

Hockney is quite well known for his works depicting swimming pools, so I entered a prompt on Dall-E, an off-chain image generator that is spiritually similar to tools like Midjourney and Stability AI. My prompt was “Woman and Man by the Pool in the Style of David Hockney.”  

Here’s what the AI generated for me: 

Clearly, the image generated does not infringe the Hockney works; the creative elements are not substantially similar. But certainly the AI captured, at least in some sense, the spirit or style of the Hockney works. You can imagine better prompts—or evolving technology—might generate images that bear an even closer resemblance to an actual Hockney painting.  

However, the artists face an uphill battle in their class action lawsuit. In large part due to the United States Supreme Court’s findings in a historic case decided a couple of years ago, Google v. Oracle (see Innovator 007 for a lengthy piece on the subject). In Google v. Oracle, the Supreme Court held that Google’s use of code copied from Java APIs (application programming interfaces) to build its Android mobile operating system was fair use.  

In other words, gathering IP and using that IP to make new IP that is “transformative” might be alright under our copyright laws. “What’s transformative?” Well, that’s another big question that is pending before the Supreme Court in Warhol v. Goldsmith (which we covered at length in Innovator 010). If an artist or creator takes an existing artwork and “transforms” it into a new, original work, then the resulting work is “fair use” and not copyright infringement. On the other hand, if an artist or creator uses existing artwork and creates a new work based on that previous artwork, but fails to “transform” the work, then the new resulting work is copyright infringement.  

When dozens of lawyers, students, professors, and clients asked me for my thoughts on this AI lawsuit, my response was twofold: I don’t think that making training data used to create new works is illegal and in violation of the Copyright Act – Google v. Oracle makes that pretty clear. Yet, I don’t think we can say that every work generated by Midjourney, Stability AI, or Deviant Art is non-infringing. The focus must be on whether the generated work is “substantially similar” to the original work. Since such an inquiry is fact sensitive and really turns on the creative elements in each specific artwork, I don’t think that this is really the kind of case that can be dealt with on a “class basis.” Rather, I think we’ll see individual artists raise claims when generated images are produced and used.  

Finally, it’s worth pointing out that the entire purpose of the Copyright Act is to inspire innovation. This country’s founders put intellectual property laws in place to encourage its citizens to be creative and make stuff. They felt that protecting creative works would incentivize people to keep creating work. In light of Congress’ intent behind the IP laws, it’s difficult to imagine a court enjoining these kinds of software tools. These tools are pathways to greater innovation, image creation, and artistry.  Any concerns that artists have with these tools are covered by the Copyright Act; a Midjourney artist must ensure that their generated artwork does not infringe anyone else’s, just like a painter must ensure that their work does not infringe anyone else’s artwork. 

This is an exciting time for AI. I completely understand, appreciate, and respect the artists’ concerns with technology running afoul of their creative property. I would encourage those concerned to see the tool as a means of exploration, not as a threat to creativity. After all, human beings work by gathering information, storing references, and recalling memory, much like the tools that are at issue in this case. People use training data every day to make new work; that we now have a tool that helps us do it should not be stopped. But those using the tools must understand that the same IP laws apply, which means that your obligations remain the same: make sure your new work doesn’t infringe anyone’s existing work.