The Next Trick for IBM's Watson? Composing Music

More than anything else, IBM's Watson supercomputer is probably best known for one thing: Appearing on Jeopardy!, the legendary TV game show, in 2011. With an internet connection and the ability the buzz in quicker than a human opponent could, Watson destroyed Jeopardy!'s longest-tenured champions, Ken Jennings and Brad Rutter, in devastating fashion throughout a week of games at the IBM campus. If you were to ask the average person if they know about IBM's supercomputer, there's no doubt that an affirmative answer would involve cleaning up on a gameshow.

There's much more to Watson, though. There are now more than 30 different Watson services as part of what IBM calls the Watson Developer Cloud, including a tool that discerns tone in writing. On top of that, there are consumer tools like Chef Watson, where the supercomputer helps generate recipes based on available ingredients.

Watson's latest frontier is a particularly human one: writing music in the form of IBM's latest project, Watson Beat. Taking some music input by humans and guidance in the form of one of four pre-programmed moods, Watson Beat will create a wholly original composition that has its own character, though you can sometimes hear the DNA of the song that it deconstructed in the final product.

Oh, and the songs will be completely different each time.

As part of the demonstration that IBM set up for Motherboard, "Happy Birthday to You," newly in the public domain, was input using a MIDI keyboard by Richard Daskas, a musician working on the project. After telling Watson Beat to run it through the "Middle Eastern" mood setting on two separate occasions, the supercomputer spat out the two compositions that are included in the above SoundCloud playlist. You can easily hear that while the input parameters were exactly the same, Watson Beat output two completely different songs.

"At the heart of Watson Beat is some deep learning and machine learning algorithms," IBM engineer Janani Mukundan explained to Motherboard. "It understands music theory. It also understands emotions, genres, and themes. It can take a piece of music, it will deconstruct it, and learn from it the essence of the piece, and also add different motions, moods, etc. on top of it."

"I was brought on the project to help guide the musical/creative process," Daska continued. [Watson] was creating random notes, and it needed to be taught music theory, just like how you would teach a child chords and rhythm, all of the elements that make music."

That's the advance in artificial intelligence here, getting a computer to think creatively.

"From the technical side of it," Mukundan explained, "there are lots of machine learning algorithms that compose what Watson Beat is doing, but at the heart of Watson Beat is an artificial neural network that tries to mimic how the brain and the central nervous system work. It understands what sounds good, what sounds bad, chord progression, emotion, and motion. If you want to hear something spooky and mysterious, it knows what that means."

So now that Watson can help you find recipes, write music, and even edit copy, more and more progress is being made each day. Whether it's helping you figure out what to throw together from what's in your refrigerator or helping you goof off in GarageBand, artificial intelligence is starting to escape the chains of its own version of the uncanny valley.