Generative AI and Copyright
Comedian Sarah Silverman is suing OpenAI and Meta – and could cause huge problems for ChatGPT
Sarah Silverman is suing ChatGPT operator OpenAI for her book “The Bedwetter”.
© Rich Fury / Getty Images
With chatbots and image creation, AI finally made it into the mainstream last year. Now the first lawsuits against ChatGPT and Co. have been filed. And that could just be the beginning.
It was a technological breakthrough that immediately made the benefits of artificial intelligence clear to everyone. With smart chatbots like ChatGPT and image creation programs like Midjourney, anyone can create text or images out of thin air in no time. As a result, AI finally made it into the mainstream last fall. But the triumph of ChatGPT and Co. could soon slow down sharply: the first celebrities are complaining.
The comedian Sarah Silverman, best known for her comedy show and numerous stand-up programs, together with the authors Christopher Golden and Richard Kadrey, has filed a lawsuit against the ChatGPT operator OpenAI and Facebook’s parent company Meta. Your accusation against the corporations: They are said to have violated copy protection laws.
reading machine
The core of the lawsuit is the way in which language AIs such as ChatGPT, so-called Large Language Models (LLM for short), are developed. Unlike classic computer programs, they are not programmed step by step by human hands, but learn independently by processing vast amounts of training material. One speaks of machine learning. And this is exactly where the lawsuit comes in: The books of the now plaintiffs are said to have been used for the training, according to the accusation. The problem: No permission was granted for it.
That is clear from the complaint. In the attached evidence, the three authors explain how extremely precisely the language AIs can summarize their books “The Bedwetter” (Silverman), “Ararat” (Golden) or “Sandman Slim” (Kadrey). Their reasoning: even if the AI makes minor mistakes, such a degree of reproduction is only possible if the language models are fed with the books. If they are now reproduced, the plaintiffs believe that this constitutes a copy protection infringement.
To make matters worse, the corporations themselves are said not to have acquired the books: In a description of his language model, Meta names a book collection known only as “The Pile” as one of his sources. The problem: The Pile was compiled from book collections from the pirate scene. And the language models are said to have used these so-called “shadow libraries” and thus violated copy protection laws even before the training began.
From Henry Cavill to Tom Hardy: AI artist shows potential Bond successors in their roles
A matter of principle
The lawsuit is explosive. Although it is known that AI models have been fed and trained with millions and millions of data sets, the operators have not yet revealed which ones exactly. Presumably also to make such lawsuits more difficult. However, if the works are used to generate new content, the question arises as to whether the creators are not entitled to a license fee.
This applies not only to written works, but also to other areas. The picture agency Getty Images is already suing Stability AI because their program Stable Fusion is said to have been trained with millions of photos from the agency – also without paying. Similar efforts are also being made in relation to the processing of program code.
Serious lawsuit
So far, the consequences of a successful lawsuit can hardly be reliably predicted. Not even in the specific case: Silverman, Golden and Kadrey have so far demanded a refund of the profits made with their materials as well as damages in an undisclosed amount. Should they win, however, the floodgates are likely to be wide open to countless more lawsuits.
In addition, the question of how AI companies deal with protected material also relates to very fundamental questions about copy protection. After all, the programs don’t do anything that people don’t do – namely learn from the works of others. However, the scale is completely different. If a court agrees with the artists and the verdict is upheld by the instances, the question arises as to whether AI can function at all in its current form. If they lose, however, that could potentially have a huge impact as well.