Meta Faces Legal Battle Over Use of Pirated Books in AI Training
In a recent legal saga, Meta, the parent company of Facebook, finds itself entangled in a lawsuit filed by a group of authors who allege that the tech giant employed copyrighted material, including the infamous Books3 dataset, in training its Llama 1 and Llama 2 large language models. While Meta has admitted to utilizing the contentious dataset, the company is hesitant to provide adequate compensation to the rights holders.
The Books3 dataset, a compilation of over 195,000 books amounting to nearly 37GB, has been a popular resource for AI training, created by AI researcher Shawn Presser in 2020. Meta’s acknowledgement of using Books3 aligns with a broader trend where major tech companies, including OpenAI and Microsoft, have faced legal challenges for incorporating copyrighted material into their AI models.
In response to the legal action, Meta asserted that its use of Books3 did not necessitate “consent, credit, or compensation” and denied any intentional misconduct. Despite acknowledging parts of the dataset’s incorporation into its LLMs, Meta maintains that it did not infringe on alleged copyrights, arguing that any unauthorized copies in Books3 fall under fair use.
Furthermore, Meta challenges the classification of the lawsuit as a Class Action, signalling reluctance to offer monetary relief to the authors involved. The company’s stance echoes broader industry sentiments, with OpenAI asserting that training AI models without copyrighted material is “impossible” and urging courts to dismiss compensation lawsuits from rights holders.
As the legal battle unfolds, Meta’s utilization of pirated books raises questions about the ethical implications of training AI models on copyrighted material and the industry’s approach to compensating rights holders. The Books3 dataset, which drew attention in 2023 when the Danish anti-piracy group Rights Alliance sought its removal, continues to be a focal point in the ongoing clash between tech companies and content creators.
