Atlantic Creates Searchable Database of Music Used in AI Training

A reporter from The Atlantic, Alex Reisner, has compiled and released a publicly searchable database detailing music datasets utilized in the training of artificial intelligence models. This initiative revealed four distinct collections, two of which are exceptionally vast, comprising 12 million and 9 million tracks respectively, alongside two smaller but still substantial sets each exceeding 100,000 songs. These datasets have seen widespread distribution, with thousands of downloads recorded. While the full extent of their usage by AI developers remains unclear, major players such as Google and Stability have acknowledged employing these resources in their research, as evidenced in their published papers. The origins of some music, like that from the Free Music Archive, permit personal streaming but restrict commercial re-use, raising questions about licensing in AI training.