Revealed: The Authors Whose Pirated Books Are Powering Generative AI (2023) (www.theatlantic.com)

A recent investigation revealed that generative AI systems, such as Meta's LLaMA, have been trained on a massive dataset of over 170,000 books, many of which were allegedly obtained through piracy. The lawsuit filed by authors Sarah Silverman, Richard Kadrey, and Christopher Golden points to copyright violations in the incorporation of their works into these AI models. This situation highlights the growing concerns surrounding intellectual property in the AI/ML community, as the very models that are reshaping communication and learning depend on unauthorized texts. The dataset known as “Books3,” which has also influenced other generative AI tools like BloombergGPT, contains a variety of fiction and nonfiction written by notable authors, underscoring the ethical implications of using copyrighted material without consent. As the debate over fair use and copyright in AI escalates, the tech industry's practices clash with the traditional publishing sector's views on intellectual property rights. The controversy raises fundamental questions about the future of AI training data and the potential monopolization of AI development by large corporations, stressing the need for clearer guidelines and respect for authors' rights in this rapidly evolving landscape.