Musicians like Backxwash and Titus Andronicus have already stumbled upon their own songs within massive, previously hidden AI training datasets. The Atlantic exposed this by publishing a searchable database of four colossal datasets used in AI development. This tool immediately confirmed what many suspected: widespread, unauthorized use of copyrighted music.
AI models devour vast amounts of music for training, yet much of this data was acquired without artists' knowledge or consent. This isn't just a technicality; it's a glaring conflict between innovation and basic intellectual property rights.
With artists already finding their work and copyright investigations underway, a torrent of legal challenges and demands for compensation from the music industry against AI developers seems inevitable. AI companies, it turns out, built their foundational models on a staggering scale of stolen content.
The Hidden Scale of AI's Music Consumption
- Two of the datasets contain 12 million and 9 million tracks, respectively, according to The Verge and Music Ally.
- Alex Reisner, an Atlantic reporter, uncovered these four datasets of music used to train AI models and made them searchable, according to MusicTech.
These staggering numbers, some reaching millions of tracks shared quietly within AI circles, reveal an industrial-scale appropriation of music. It's clear AI companies prioritized speed over legality, incurring significant retroactive liability.
A Precedent of Unsanctioned Data Use
Google, for instance, reportedly downloaded a smaller dataset from the Free Music Archive to train its AI models, according to MusicTech. This isn't just about massive data grabs; it's a pattern of major tech companies leveraging even 'open' archives without clear consent. This practice blurs the lines for all music data, making future licensing models unnecessarily complex.
