The Atlantic Builds Searchable Database of AI Training Music
▼ Summary
– Alex Reisner uncovered four datasets of music used to train AI models, with two containing 12 million and 9 million tracks and two others each over 100,000 songs.
– Google and Stability have confirmed using these datasets in research papers, and the sets have been downloaded thousands of times.
– Some dataset sources, like the Free Music Archive, are free for personal streaming but require licensing for commercial use.
– Three datasets are distributed as lists of YouTube or Spotify links, and AI developers use automated tools to download the audio.
– These automated tools can bypass logins, ads, and revenue mechanisms, violating the platforms’ terms of service.
The Atlantic has launched a searchable database that catalogs music used to train artificial intelligence systems, giving the public an unprecedented look into the datasets powering modern AI music generation. Reporter Alex Reisner identified four distinct collections, two of which are massive in scale: one contains 12 million tracks, while another holds 9 million songs. Two smaller datasets each include over 100,000 pieces of music, still representing a substantial volume of training material.
These datasets have been downloaded thousands of times, according to Reisner. Although it remains unclear exactly which organizations have accessed them, both Google and Stability AI have acknowledged using the data in published research papers. The legal landscape is complicated: some sources, such as the Free Music Archive dataset, allow free streaming for personal listening but require commercial licensing for any corporate or AI training use.
Accessing these datasets for AI training is not as straightforward as downloading a single file. Reisner explains that three of the collections are distributed as links pointing to songs on YouTube or Spotify. Developers then use automated tools to extract the actual audio, often bypassing login screens, advertisements, and revenue-generating features for creators. These methods violate the platforms’ terms of service, raising further questions about the ethical and legal boundaries of AI training practices.
(Source: The Verge)
