Apple trained AI models on YouTube content without consent

A number of tech giants, including Apple, trained AI models on YouTube videos without the consent of the creators, according to a new report today.

They did this by using subtitle files downloaded by a third party from more than 170,000 videos. Creators affected include tech reviewer Marquees Brownlee (MKBHD), MrBeast, PewDiePie, Stephen Colbert, John Oliver, and Jimmy Kimmel …

The subtitle files are effectively transcripts of the video content.

Wired reports.

An investigation by Proof News found some of the wealthiest AI companies in the world have used material from thousands of YouTube videos to train AI. Companies did so despite YouTube’s rules against harvesting materials from the platform without permission.

Our investigation found that subtitles from 173,536 YouTube videos, siphoned from more than 48,000 channels, were used by Silicon Valley heavyweights, including Anthropic, Nvidia, Apple, and Salesforce.

The downloads were reportedly performed by a non-profit called EleutherAI, which says it helps developers train AI models. While the aim appears to have been to provide training materials to small developers and academics, the dataset has also been used by several tech giants, including Apple.

According to a research paper published by EleutherAI, the dataset is part of a compilation the nonprofit released called the Pile […]

Most of the Pile’s datasets are accessible and open for anyone on the internet with enough space and computing power to access them. Academics and other developers outside of Big Tech made use of the dataset, but they weren’t the only ones.

Apple, Nvidia, and Salesforce—companies valued in the hundreds of billions and trillions of dollars—describe in their research papers and posts how they used the Pile to train AI. Documents also show Apple used the Pile to train OpenELM, a high-profile model released in April, weeks before the company revealed it will add new AI capabilities to iPhones and MacBooks.

Wired says Apple hadn’t responded to a request for comment at the time of writing.

9to5Mac’s Take

It’s important to emphasize here that Apple didn’t download the data itself, but this was instead performed by EleutherAI. It is this organization which appears to have broken YouTube’s terms and conditions.

All the same, while Apple and the other companies named likely used a publicly-available dataset in good faith, it’s a good illustration of the legal minefield created by scraping the web to train AI systems. There have been multiple examples of AI systems plagiarizing entire paragraphs of text when asked about niche topics, and the dangers of using material without permission are only increased when companies use datasets compiled by third parties.

We’ve reached out to Apple for comment, and will update with any response.

Screengrab: MKBHD