Copyright and artificial intelligence: innovation or infringement?

By Jean G. Vidal-Font, Senior Member Attorney at Ferraiuoli LLC

ChatGPT was the first platform made available to the general public whose interface was natural and accessible enough for anyone to use. Soon after, new platforms came out aimed at producing images created by AI using only basic instructions from users.

One of the most popular has been Stability.AI, which generates new photos based on a sample that the user uploads. For example, the user uploads several profile photos and Stability.AI generates countless new photos with different themes (cowboy, astronaut, anime, etc.).

But how do platforms like Stability.AI learn to generate new works and photographs?

According to a lawsuit filed by Getty Images, the platform learns at the expense of third-party art (Getty Images Inc v. Stability AI Inc, U.S. District Court for the District of Delaware, No. 1:23-cv-00135). In a current case at a Delaware Federal Court, Getty Images filed a copyright infringement lawsuit against Stability.AI, alleging that the platform "learns" to generate new images by capturing and reviewing pre-existing images, such as the Getty Images photobase. Getty Images argues that the artificial intelligence company misused more than 12 million Getty Images photos to train its Stable Diffusion AI image-generation system (1).

A similar allegation is found in the class action filed in a Court of California (Andersen v. Stability AI Ltd, U.S. District Court for the Northern District of California, No. 3:23-cv-00201).

Similar to the Getty Images case, the allegation focuses on the fact that Stability.AI uses the images of others to train its algorithm, and therefore, although the final work generated by the platform could be original, it can only be possible through the reproduction and derivative use of the works of others, without authorization.

Both cases are in the early stages of litigation, but how they conclude may be crucial for the progress of the next generation of AI for imaging.

As a general rule, no one can use the work of another without their permission, this being the basic rule behind copyright, either in its federal modality, under the Copyright Act, or its local modality in Puerto Rico, under the Law of Moral Rights.

But to what extent does the use of an image for "educational" purposes by an algorithm constitute an infringement?

Although the AI-generated work is not a properly derivative work of the original work, the AI-generated work would not be possible without the use of the original work.

Content creators and photographers, among others, are entering a new world where AI may replace them but at the same time depends on their pre-existing universe of images. A platform like Stability.AI might not be possible if you had to pay licenses or royalties for the photos you use, not to reproduce or adapt, but to process and learn.

Are we facing a case where current law prevents technological progress? Or are we facing a similar case to Napster, where at the end of the day, it was the technological and business model that was adapted to the applicable legislation?

Copyrights are there to encourage the creation of new arts and expressions, which will only be possible if artists feel safe at protecting their rights.

This article was originally published in Spanish in Microjuris on April 29, 2023