A federal judge ruled in a landmark case for the artificial intelligence sector that Humanity, an AI firm, did not violate the law by training its chatbot models on millions of copyrighted texts.
However, the company still faces legal challenges over the methods used to acquire these texts, specifically downloading them from an online “Shadow Library” filled with pirated materials.
U.S. District Judge William Alsup stated in a ruling issued late Monday that this constitutes a “typical transformation” process, as AI systems extract insights from numerous written works, enabling them to generate original content that is considered “fair use” under U.S. copyright legislation.
“Like aspiring writers who critique existing literature, Humanity (the AI language model) trained on these works without aiming to replicate or replace them initially.”
Despite dismissing key claims brought by a group of authors who filed a copyright infringement suit against the company last year, Alsup noted that Humanity still must stand trial in December for allegations of theft regarding their works.
“Humanity had no right to utilize pirated copies in the Central Library,” Judge Alsup stated.
Last summer, authors Andrea Burtz, Charles Graeber, and Kirk Wallace Johnson contended that Humanity’s practices amounted to “massive theft,” asserting that the San Francisco-based company “aimed to profit from the creative talent and expression underlying these works.”
Books are recognized as a crucial resource (essentially billions of words meticulously curated) required for developing large-scale language models. In the competitive arena of crafting cutting-edge AI chatbots, many tech companies are turning to freely available online repositories of stolen texts.
Documents revealed in federal court filings from San Francisco indicate that internal concerns existed among Human employees regarding the legality of utilizing pirate sites. The company subsequently revised its strategy and appointed former Google executive Tom Turvey to lead the efforts involving Google Books. Years of copyright contention.
According to court documents, Humanity, with Turvey’s assistance, procured a substantial number of books, disassembled them, and began scanning each page to feed the data into an AI model. However, the judge stated that this did not absolve prior copyright infringements.
“The fact that Humanity later obtained legitimate copies of books originally pilfered from the Internet does not nullify liability for theft; however, it could influence the extent of statutory damages,” wrote Alsup.
A judgment could set a precedent for similar lawsuits. Other entities like Anthropic and OpenAI, which created ChatGPT, are also implicated. Meta Platforms, the parent company of Facebook and Instagram, is under scrutiny as well.
Founded in 2021 by a former OpenAI leader, Humanity markets itself as a creator of a more responsible and secure generative AI model designed for tasks like composing emails, summarizing documents, and interacting with users in a natural manner.
Nonetheless, a lawsuit filed last year argued that Humanity’s actions “disregarded the noble objectives” by integrating AI products with pirated materials.
On Tuesday, Humanity stated that the judge acknowledged the compatibility of AI training with the “copyright purpose that fosters creativity and encourages scientific progress.” The statement did not address any allegations of copyright infringement.
The plaintiffs’ attorney declined to provide comments.
Source: apnews.com