The AI industry was rather dull and boring.
But since OpenAI introduced ChatGPT, the company created an arms race, where companies of all sizes either develop their own AI tools, or use others' tools, in order to enhance the experience of using their products.
With tech companies going neck-and-neck, their AI tools hunger for more data.
To satisfy their hunger, they have literally depleted the information they can get from the text-based web.
Moving forward, AI companies are training their AI on videos. And on the internet, one of the largest sources of videos, is YouTube.
And here, tech companies are found training their data on YouTube videos, without creators' or even Google's consent.
Back earlier in April, YouTube sent a clear message to AI model developers that downloading data from the platform and using it to train AI models is a clear violation of its terms of service.
A Google spokesperson further clarified this, saying that any, "unauthorized scraping or downloading of YouTube content" is prohibited.
But according to a report from Proof News, YouTube has been scraped for its data, and some of the biggest tech companies advancing AI have used it to train models.
According to the nonprofit news studio in a post on its website, subtitles from 172,535 YouTube videos were siphoned from more than 48,000 channels, and some of these channels included prominent creators on the platform.
They include MKBHD (19 million subscribers), MrBeast (289 million), Jacksepticeye (31 million), PewDiePie (111 million), and more.
Educational and online learning channels like Khan Academy, MIT, and Harvard, news agencies like The Wall Street Journal, NPR, and the BBC also had their videos used to train AI, and so did The Late Show With Stephen Colbert, Last Week Tonight With John Oliver, and Jimmy Kimmel Live.
Some of the material used to train AI also promoted conspiracies such as the "flat-Earth theory."
The dataset that compiled information from these channels, is called YouTube Subtitles, which doesn't include video imagery but consists of plain text of videos’ subtitles, often along with translations into languages including Japanese, German, and Arabic.
And as for who is scraping these channels' data, Proof News said that they include Silicon Valley heavyweights, including Anthropic, Nvidia, Apple, and Salesforce.
"AI companies are generally secretive about their sources of training data, but 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," Proof News said.
According to a research paper published by EleutherAI, the dataset is part of a compilation called the Pile.
Pile, which is a 800GB dataset of "diverse text for language modeling," included material from not just YouTube, but also the European Parliament, English Wikipedia, and a trove of Enron Corporation employees’ emails that was released as part of a federal investigation into the firm.
Most of the Pile’s datasets are accessible and open for anyone on the internet with enough space and computing power to access it.
The creator of the dataset, called EleutherAI, states that its overall goal is to lower the barriers to AI development to those outside the walls of big tech companies.
In the past, EleutherAI claimed to have provided "access to cutting‑edge AI technologies by training and releasing models."
It should be noted that all of the big tech companies listed above didn't download the YouTube video transcriptions.
Allegedly, the companies discovered the dataset and decided to use it to train their AI models, which raises the question of what happens when a company uses a dataset from a third party to train an AI model, but that dataset contains data that users didn't consent to be used for training purposes.
In Apple's case, the Pile dataset was used to train OpenELM, an AI model that was released weeks before the company revealed Apple Intelligence.
