The court decision that could turn generative AI upside down | Technology

One of the biggest unsolved problems surrounding generative artificial intelligence is copyright. In order to function, these tools need to process billions of texts, images or videos, from which they then extract patterns that allow them to create seemingly original content. Many creators and artists complain that the companies responsible for the AI ​​models use their work without their consent, or even that they recognise their personal style in some of the creations of these models.

Between 2022 and 2023, a group of programmers, writers, screenwriters and artists filed four major class action lawsuits (class actionsin English) against the major developers of generative AI, including Meta, Microsoft and OpenAI. One of these processes, initiated by several illustrators, obtained a significant victory two weeks ago: Northern District of California Judge William H. Orrick dismissed the allegations of the companies accused — Stability AI, responsible for Stable Diffusion; Runway AI; Deviant Art, and Midjourney — and admitted the plaintiffs’ main claims.

This means that the case is moving forward and will enter the so-called discovery phase. “Now is when we are allowed to request documents from the defendants and take statements or testimony. We will ask the companies that trained the AI ​​image generation models to provide information about how they copied the plaintiffs’ work and how they have used it in the development of their tools,” Matthew Butterick, one of the lawyers handling this lawsuit, explains to EL PAÍS.

Puerto Rican illustrator Karla Ortiz, one of the plaintiffs, does not hide her euphoria. “We are now potentially one of the largest copyright infringement cases in history. We are excited for the next stage of our fight!” she said on social media as soon as she heard the news. Ortiz, whose brushes have helped shape characters in blockbusters such as Thor: Ragnarok, Doctor Strange either Jurassic World and a board member of the Conceptual Art Association of America, she has emerged as one of the greatest defenders of her profession in the face of the AI ​​wave. “My job is at risk. It consists of showing ideas, and machines do that very well now. Artists cannot compete against these tools. I had never worried about the future of my career until now,” she told this newspaper earlier this year.

Rodrigo Cetina, a law professor at the Barcelona School of Management, the business school of the Pompeu Fabra University, is an expert on the American legal system. He believes that the fact that the lawsuit has been admitted for processing “is a sign that the judge believes that the answer to the key questions is probably affirmative: whether copyright is infringed in the training of an AI and whether it is an infringement to copy billions of images from the internet to train your model.”

More specifically, the judge has agreed to assess the possible copyright infringement of Stability AI, Runway AI and Deviant Art and fraudulent use by Midjourney of the names and distinctive visual styles of the plaintiff artists. To decide whether or not the companies have to compensate the illustrators, the judge will have to establish whether their works have effectively been copied or, at least, pass an acceptable threshold of copying, in ways that cannot be considered fair use, Cetina said.

What exactly will the judge look at to decide whether the copying of the works was significant or not? “Generally, a test developed in jurisprudence is applied that considers four factors: the nature of the protected work, the purpose for which it is used, how substantial the proportion of a work used is and the effect of the use in question on the market potential for the protected works,” Cetina explains. It is a complex process, and it is difficult to anticipate how the judge will assess the different factors.

Although Californian judges, this legal expert recalls, tend to be protectionist towards creative industries, as happened for example with the Napster case, the online music file-sharing portal that was closed by court decision in 2001. “Something very important in the substantive copy test is whether there has been access to previous works and whether there is a high degree of similarity between the original and the supposedly infringing work. So, if the work generated by AI is sufficiently similar, there could be something there,” adds Cetina.

The future of AI lies in the courts

The potential of generative AI became known to the general public in November 2022, when OpenAI presented its flagship tool: ChatGPT. Suddenly, we could converse with a machine that seemed to understand us, that answered our questions with ease, that was able to follow a conversation and that, after a few months in which it lied more than it talked, proved to be relatively reliable.

That was just the beginning. Other tools soon appeared, such as Dall-E, Stable Diffusion or Midjourney, capable of creating sophisticated and realistic images from a series of written instructions. The latest to arrive have been hyperrealistic video generators, such as Sora, from OpenAI, whose potential is both technically fascinating and ethically terrifying.

Some people sensed danger from the start, when the world was in full awe of the possibilities offered by this new technology. Butterick did so in the summer of 2022, before the arrival of ChatGPT. He was alerted by the launch of Microsoft’s GitHub Copilot, an AI-assisted programming tool that is trained with a large amount of open source software. The American, who is a programmer as well as a lawyer, filed a still-unresolved lawsuit against Microsoft in November of that year, accusing it of violating open license agreements.

That was the first legal attack against generative AI. In January 2023, the illustrators’ attack would come, and there has just been news about it. In July, it was the turn of a group of writers, who sued OpenAI and Meta for including books written by them in their training databases. In October last year, several record labels, including Universal Music Group, sued OpenAI and Meta. Anthropic for training their algorithms with copyrighted materials.

Since then, lawsuits have multiplied: Getty Images sued Stability AI for using images from its archives without permission, The New York Times took OpenAI and Microsoft to court for using millions of articles in training ChatGPT, and more writers (including George RR Martin and Jonathan Franzen) filed lawsuits against OpenAI on the same grounds.

You can follow THE COUNTRY Technology in Facebook and X or sign up here to receive our weekly newsletter.