Back to the Future …of AI
A time experiment on the use of AI in generating images. Who (or what) is the Creator of an AI Image?
A fundamental element of copyright law is that for a work to be eligible for copyright protection it must be created by a human.
Above are examples of different output images created using Midjourney AI from identical text prompts separated by time. The first image is a recent creation using Midjourney V6(Alpha) and the accompanying image was created at some point in 2022 using a prior version of Midjourney. List of prompts: (i) Blue ridge mountains with infinite clouds and farmland with a red barn and silo; (ii) An infinite gothic catholic church with heaven ceiling and monks; and (iii) Green landscape endless field with fireflies and trees with mountains in background at twilight.
From the above examples it is clear the style and resulting image vary widely. The “controls” include the text prompts themselves, which were identical, as well as using the same Midjourney account to create both versions. There were variables, including the technology itself, which advanced over time, as well as the default settings. It seems clear that Midjourney’s image generation process takes into account your history of image generation on your specific account, the millions of prompts and images generated from the many users of the technology, as well as sourcing and “training” itself on a multitude of existing human generated works (many of which have not entered the public domain). Similar to search algorithms, the “secret sauce” of image generation and the process used by the platform is largely a black box.
Théâtre D'opéra Spatial. 2022.
Artist: Midjourney and Jason Allen. Generated using Midjourney AI in response to a series of prompts provided by Jason Allen.
Jason Allen sought copyright protection of the above work titled “Théâtre D'opéra Spatial” claiming that he created the image using Midjourney after multiple iterations of prompts fine tuning the output and then revising using Adobe Photoshop. The Copyright office denied registration because Mr. Allen refused to disclaim those portions of the work that were generated by Midjourney, which the Copyright Office determined to be the product of “non-human” creators. This decision was contested by Mr. Allen and upheld by the Copyright Office, even with Mr. Allen’s claim regarding the significant human input needed in developing the desired output by revising the text prompts.
In a world of computer aided applications and generative artificial intelligence, when do we cross the line that will preclude protection of the resulting works under US copyright law? Can anyone claim ownership and exclusive usage of AI generated works, and can the process of “training” AI infringe on the rights of those copyright owners whose works the AI sourced for such training?
The US Copyright Office has issued a Notice of Inquiry and Request for Comments in evaluating policy related to Artificial Intelligence and the interaction with Copyright Law. Over 10,000 public comments were submitted in response to the request. Unsurprisingly, there are multiple lawsuits surrounding generative AI, including one recent lawsuit for copyright infringement brought by The New York Times against Microsoft and OpenAI alleging copyright infringement in using The New York Times content to train its AI products and services. In many ways it’s fitting that an institution that had reporters on the front line of the Civil War could well be the catalyst to the courts addressing a number of fundamental legal questions involving AI.
As we go into 2024, we can expect generative AI to continue to develop and we should also start to see more specific guidance and rulings on how to thread the needle on the “fair use” legal doctrine in copyright law in the time of AI, as well as better defining a human’s contribution and when the use of technology transfers the creative process and resulting ownership of works from the human individual to the “other” that is AI.