Authors Guild Urges AI Companies to Compensate Writers for Training Models on Their Work
More than 8,000 writers have signed an open letter from the US Authors Guild, calling on leaders of six major AI companies to obtain consent and fairly compensate authors for using their copyrighted works to train AI models.
Large language models rely on vast amounts of scraped text from the internet for training. However, numerous books available on websites have been ingested without the authors’ permission. Many writers are now demanding that their work no longer be exploited by computers.
The letter, addressed to the CEOs of OpenAI, Alphabet, Stability AI, Meta, IBM, and Microsoft, highlights that generative AI technologies owe their existence to authors’ writings. These technologies imitate and reproduce language, stories, style, and ideas originating from millions of copyrighted books, articles, essays, and poetry.
The Authors Guild argues that since these companies invest billions of dollars in developing AI technology, it is only fair for them to compensate authors for using their writings. Without these creative works, AI would be unexceptional and severely limited.
Mary Rasenberger, CEO of the Authors Guild, explained to NPR that the letter aims to encourage settlements between companies and writers without resorting to legal action, which can be costly and time-consuming. Nonetheless, some writers have taken a more aggressive approach by filing lawsuits against those they believe have stolen their work.
In an effort to compete with rival AI developers, Meta is reportedly launching an updated version of its large language model, LLaMA, which will support commercial use. Meta has often released its models for academic research but faced criticism for being less open than claimed, denying developers the ability to use LLaMA for commercial applications.
According to the Financial Times, Meta is exploring options to potentially charge enterprises for fine-tuning the model using their own custom data. However, the company may ultimately decide not to charge users at all. By sharing its models with more developers, Meta hopes to diminish the shine of competitors like OpenAI, Google, and Microsoft. The rumor regarding Meta’s intentions was first reported by The Information last month.
To address a wave of copyright lawsuits, OpenAI has recently partnered with the Associated Press (AP) and Shutterstock to license their content for training AI models. While this move provides limited relief for thousands of book authors and millions of authors of other written works, OpenAI’s partnerships serve a specific purpose.
Under the agreement, AP will grant OpenAI access to an archive of text dating back to 1985, and OpenAI will provide the non-profit news agency with technology and product expertise. The collaboration aims to explore potential use cases for generative AI in news products and services. Similarly, Shutterstock signed a six-year agreement with OpenAI allowing the company to license its content, empowering OpenAI in training its generative AI systems, while Shutterstock can continue utilizing its technology to enhance tools like the AI Image Generator. Financial details of both deals remain undisclosed.
By adhering to these guidelines, AI companies address the concerns raised by the Authors Guild and work towards ensuring fair compensation for writers whose works are instrumental in training AI models. As the debate surrounding copyright infringement and AI technology continues, open dialogue and mutually beneficial solutions become essential.