The Oregon-based artist Sarah Andersen draws comics entitled “Sarah’s Scribbles,” and she’s sick and tired of artificial intelligence (AI) using her material without permission or compensation.

Along with two other artists, Andersen filed a class-action lawsuit in federal court in San Francisco last month, accusing Stability AI, Midjourney, and DeviantArt—leading AI developers—of infringing the plaintiffs’ copyrights by running their protected artwork through “training data” that the companies’ AI uses and/or by generating “derivative” works from that training set.

Labeling AI image generators such as Midjourney and Stable Diffusion “21st century collage tools that violate the rights of millions of artists,” the plaintiffs contend in their complaint that they seek to curtail robotic art makers’ infringement “before their professions are eliminated by a computer program powered entirely by their hard work.” Before now, purchasers would compensate artists via licensure or other methods that enabled use of their works. Now all they have to do is ask Stable Diffusion to “generate new works in the artist’s style without compensating the artist at all.”

Not to be outdone, Getty Images, one of the world’s leading repositories of stock photos and other content, filed suit in London that same week against the UK-based Stability AI, asserting in a press release that Stability AI “unlawfully copied and processed millions of images protected by copyright and the associated metadata owned or represented by Getty Images absent a license to benefit Stability AI’s commercial interests and to the detriment of the content creators.”

Are AI companies allowed to use copyrighted works for their training sets, and can their programs generate further works “in the style” of known artists? Should they be able to do so?

The main defense that Stability AI and its allies will likely mount involves the complex copyright doctrine of fair use. Under that doctrine, as we explored in this space in the context of a software suit between Oracle and Google, courts typically consider whether the use of a copyrighted work is transformative, how creative the copyrighted work is, how much of the work the infringer used, and how much harm the infringing use caused.

Here, the artists specifically accuse the defendants of infringement both through use of the artwork in their training datasets and through creating new works on the basis of that training. In both situations, the companies appear to be using copyrighted work that is truly and fully creative. But the rest of the analysis differs according to the accused type of use.

With respect to the “input” (i.e., using copyrighted works to train AI), there doesn’t appear to be much transformation of the works, which appear to be used in their entirety; by all accounts, the programs ingest the works of art wholesale and do not transform them. However, it will likely be difficult for the artists to prove that they’ve been harmed by the mere training of an AI model on their works.

With respect to the “output” (i.e., using the training data to generate new works “in the style” of the underlying art), the harm experienced by the artists may indeed be profound, especially if they can prove that the market for their “real,” human works has been harmed by the cheap (or free) availability of AI-produced facsimiles. What remains unclear is how much of the artists’ work is actually being used in the AI-generated images and how transformative those images actually are.

The defendants certainly know where they stand on these issues. “Anyone that believes that this isn’t fair use does not understand the technology and misunderstands the law,” a Stability AI spokesperson told Bloomberg.

But the policy implications of the lawsuit are even more significant. Would a finding in favor of the artists impede the development of AI? One prominent copyright scholar thinks so, cautioning that finding in favor of the artists “would just slam the doors on the research, it would slam the doors on certainly the applications that Stability AI has created and DALL-E, and all the other products that follow a very similar formula.”

But perhaps a middle ground is available. According to Bloomberg, the artists’ lawyers argue in favor of an art-licensing system, akin to Apple Music or Spotify, that would compensate artists for the use of their images in training sets and in the artwork that emerges in their style.

For its part, Getty said in its press release that it “believes artificial intelligence has the potential to stimulate creative endeavors” and therefore “provided licenses to leading technology innovators for purposes related to training artificial intelligence systems.” But, according to Getty, “Stability AI did not seek any such license” and instead “chose to ignore viable licensing options and long‑standing legal protections in pursuit of their stand‑alone commercial interests.”

How these suits play out may well chart the course for the future development of AI.

The post Can AI Violate Copyright? A New Lawsuit Argues Yes. appeared first on American Enterprise Institute – AEI.