DeepMind Developed… and Shelved… an AI Music Generator

Google’s DeepMind unit created an artist-mimicking AI music generator, codenamed “Orca,” that allegedly worked so well the tech giant couldn’t release it for fear of legal fallout.

Why It Hits: It’s no secret that AI systems have been trained on vast amounts of copyrighted data. It’s what makes them so powerful… and a disaster when it comes to the law. Knowing that Google is already capable of building a system that can mimic an artist may inspire the creative community to push for wide-ranging protections before Pandora’s box is opened.

Between the Code: Google’s Orca could potentially do as much damage to the music industry as real Orcas can do to small boats… at least, according to four people involved in the project who anonymously told Insider.

• DeepMind and YouTube collaborated on the tool, which explains where all the premium copyrighted data came from.
• The tool allowed users to generate an entire song by simply prompting it with a specific artist, a genre, and some lyrics… and the results were, according to a source, “mind-blowing.”
• Google floated the idea of releasing it publicly to a few music labels, dangling the carrot of a revenue-sharing agreement with artists whose work Orca was trained on.
• The labels swiftly shut the plan down, saying, “You’ll be hearing from our lawyers” (or, at least, we can safely assume).
• Recognizing that the tool could pose a “huge legal risk,” Google backtracked.

Final Note: Although Orca was never released, it did provide the basis for DeepMind’s other AI music generator, Lyria. Announced last year, Lyria could only generate part of a song in the style of a few artists who partnered with Google on the tech. 

So, why did Google (and specifically YouTube) consider releasing this tool even though it could so clearly hurt their standing with artists and labels? Turns out, OpenAI’s blatant scraping of YouTube content prompted Google execs to ask, “Hey, if they’re doing it to us, why not do it to ourselves?” Move fast and break things is still clearly the law of Silicon Valley.