The Atlantic created a searchable database of the music used to train AI

The Atlantic has released a searchable database that highlights how music is used to train artificial intelligence, revealing vast collections powering AI research. The four datasets include two enormous sets with 12 million and 9 million tracks, and two smaller ones above 100,000 songs each. These datasets have been downloaded thousands of times, with Google and Stability cited in research papers, though licensing and authorship vary by source. Free Music Archive, for example, allows personal streaming but requires licensing for commercial use. Many sources link to songs on YouTube or Spotify rather than providing direct audio files.

Training often requires downloading audio via automated tools that can bypass logins and ads, potentially violating terms of service. The datasets span artists from Lady Gaga and Radiohead to Bruce Springsteen and Wu-Tang Clan, illustrating the breadth of materials used to train models. The Atlantic maintains an AI Watchdog site where readers can search the referenced media.