TLDR: A new investigation by The Atlantic’s ‘AI Watchdog’ subsite has uncovered that over 15.8 million YouTube videos from more than 2 million channels were downloaded without permission to train generative AI models. A searchable database is reportedly available for creators to check if their content was used.
The Atlantic has published a significant investigation, shedding light on the extensive and unauthorized use of YouTube videos for training generative artificial intelligence models. This comprehensive report, featured on The Atlantic’s new ‘AI Watchdog’ subsite, highlights a growing concern among content creators regarding intellectual property rights in the rapidly evolving landscape of AI development.
The investigation’s key findings reveal that a staggering 15.8 million videos, sourced from over 2 million YouTube channels, were downloaded without explicit permission from their creators. These vast datasets are being actively utilized by various technology companies to train their advanced generative AI systems, raising questions about ethical data sourcing and copyright infringement.
This report underscores the profound implications for filmmakers and content creators. Their original work, often the product of significant time, resources, and financial investment, is being leveraged to develop AI programs that could potentially compete with or even replace human creative efforts. As one related article noted, ‘The companies behind the scraping are not like upstarts; they’re huge corporations using the stuff you put on YouTube to train the programs they want to replace you.’
In a move towards greater transparency and accountability, The Atlantic has reportedly made a searchable database available to the public. This tool allows individual creators to determine if their specific videos have been included in these AI training datasets and to identify the tech companies responsible for utilizing their material. This initiative aims to provide creators with much-needed visibility into how their content is being consumed by AI developers.
Also Read:
- Sweden Pioneers Global AI Music Licensing to Safeguard Songwriters’ Rights
- Konami Pulls Yu-Gi-Oh! Event Videos Over Uncredited AI Voice-Overs Resembling Popular Actress
This large-scale scraping by major tech entities is not an isolated incident. A separate investigation by Proof News, in collaboration with Wired, published on July 17, 2024, also highlighted similar practices. That report indicated that tech giants such as NVIDIA, Apple, Salesforce, and Anthropic utilized subtitles from over 173,536 YouTube videos, sourced from 48,000 channels, to train their AI models, allegedly in contravention of YouTube’s terms of service. Quotes from creators in that report, such as Nebula CEO Dave Wiskus stating ‘It’s theft,’ and David Pakman emphasizing, ‘No one came to me and said, ‘We would like to use this.’ This is my livelihood, and I put time, resources, money, and staff time into creating this content,’ reflect the widespread frustration and sense of violation among content creators regarding the unauthorized appropriation of their work for AI training.
