Artificial intelligence (AI) models have become crucial in various industries. These models rely heavily on vast amounts of data to function effectively. However, an investigation by Proof News and Wired has revealed that major AI developers, including Apple, Nvidia, and Anthropic, are using transcribed YouTube videos without the creators’ permission, in violation of YouTube’s own rules.
Researchers discovered that these AI firms have trained their models with a dataset called YouTube Subtitles, which incorporates transcripts from almost 175,000 videos across 48,000 channels. The creators of these videos are completely unaware that their content is being used in this manner. The YouTube Subtitles dataset, developed by EleutherAI, aims to democratize access to AI development for individuals outside big tech companies. It is just one component of the larger EleutherAI dataset called the Pile, which also includes Wikipedia articles, European Parliament speeches, and even emails from Enron.
The Pile has gained popularity among major tech companies. For instance, Apple has utilized the Pile to train its OpenELM AI model, while Salesforce’s AI model, released two years ago, has been downloaded over 86,000 times. The YouTube Subtitles dataset includes content from a range of popular channels across news, education, and entertainment, including videos from prominent YouTubers like MrBeast and Marques Brownlee. Proof News has even created a search tool to determine whether specific videos or channels are included in the dataset.
The use of the YouTube Subtitles dataset appears to contradict YouTube’s terms of service, which explicitly prohibit automated scraping of videos and associated data. However, the dataset was created by downloading subtitles through YouTube’s API using a script. The investigation revealed that the automated download targeted videos containing nearly 500 specific search terms.
Upon learning about the unauthorized use of their content, many YouTube creators expressed surprise and anger. It is understandable that they would be upset at the idea of their work being utilized without permission or payment in AI models. Some creators were particularly disheartened to discover that the dataset contains transcripts of their deleted videos or comes from creators who have since removed their online presence entirely.
EleutherAI, the organization responsible for the creation of the dataset, did not provide any comments on the matter. The organization describes its mission as democratizing access to AI technologies by releasing trained models. However, such objectives may clash with the interests of content creators and platforms. This revelation is likely to complicate the already complex legal and regulatory battles surrounding AI development. It will undoubtedly intensify discussions surrounding the ethical and legal landscapes of AI, making it challenging to strike a balance between innovation and responsibility.
Moving forward, it is crucial for AI developers to ensure that they obtain proper permissions and follow the guidelines set by content creators and platforms. Mutual collaboration and consent will not only protect the rights of creators, but also foster a more ethical and responsible AI ecosystem. Furthermore, organizations like EleutherAI should be transparent about their data collection methodologies and seek ethical approval when using copyrighted materials. By promoting ethical practices and maintaining open lines of communication, the AI community can navigate the challenges of data acquisition, while promoting innovation and protecting the rights of content creators.
Source link
