Google AI produces a melody that rivals the Casio keyboard concerts of our youth

It’s 1989. Your parents are watching the Berlin Wall being torn down, but you’re upstairs in your room, fresh batteries in your Casio SA-1, prepping for your concert. The built-in beats are hot. You’ve been listening to the “Wake me up before you go-go” demo song and you think you’ve been inspired. You select instrument number 02, “HONKY-TONK PIANO.” It’s time to tickle the plastic ivories and burn this mother down. Hit it!

The beat drops here ^

Pretty good, right? But what if I told you that this was not an obsessively practiced performance by a precocious 7-year-old, but the original creation of an artistic artificial intelligence, the work of the boffins at Google Research (specifically, and naturally, Google Brain)? You probably wouldn’t be surprised, because that’s in the headline. But imagine the impact if you’d skipped the headline and went straight to the body of the article (😂).

Yes, this is the first public output of Magenta, “a research project to advance the state of the art in machine intelligence for music and art generation.” Machine learning, suggests Douglas Eck, one of the project’s scientists, has come a long way when it comes to speech recognition and other, more straightforward tasks. But can it learn what it is that makes a piece of music listenable, or a piece of art perusable? (We already know it can do very strange poetry (PDF).)

“We believe that the models that have worked so well in speech recognition, translation and image annotation will seed an exciting new crop of tools for art and music creation,” wrote Eck.

Like any other creative entity, Magenta needs to get its work out there, and it needs feedback. To that end, Google is soliciting creatives and coders both to join the community, check out the code, feed it data and so on. The project will live on both GitHub and use Google’s own open-source machine learning platform TensorFlow. So get in there and fork it.

It’s far from the first time researchers have looked at the possibility of computer-generated music (research stretches back decades), but with the resources and brainpower of Google Research behind it, Magenta could be one of the more interesting efforts to do so.

You can follow along with the team’s progress at the (currently minimal) Magenta blog.