By Jeremy Cook
In a 2018 article, we explored how an ATmega328P microcontroller onboard an Arduino Uno can be used to produce very rudimentary music from a small speaker. As outlined here, the ATtiny85 can even produce tones with a bit of clever programming. At the end of the day, however, all these solutions mean that your controller is repeating human inputs, and couldn’t be considered a music robot, per se.
One might suppose (and even hope) that the musical genius of Mozart, Beethoven, and other long-past artists could never be replicated by a music robot or AI musician. But with advances in computing technology and machine learning, this may actually be possible.
Some human musicians are skilled at composing songs, while others specialize in playing back melodies that others have written. And some, such as jazz musicians and jam bands, specialize in making up music on-the-fly.
There’s a huge amount of overlap between these disciplines in human musicians. Robots and automation, generally speaking, tend to focus on a singular task. That doesn’t mean they can’t learn. While improvisational robot bands aren’t yet commonplace, they do exist in several forms today.
One Love Machine Band plays MIDI-programmed tunes
As featured on CNN, Kolja Kugler’s One Love Machine Band is able to play the bass guitar and drum kit via a pair of vaguely humanoid robots. Mechanical birds whistle along for accompaniment, echoing automation concepts of the ancient past.
Materials for these particular robots are salvaged from junkyards, and movement power is supplied by compressed air. While these robots are playing pre-programmed melodies, one might say that the real art here is the creation of these eclectic beings. Kugler notes that they have “an affinity for punk rock,” though their visual style also evokes a bit of a reggae feel.
Note that while in some ways similar to automations performing at rodent-themed children's restaurants or amusement parks, One Love Machine Band actually plays their instruments via MIDI programming. You might say that One Love consists of actual robo-musicians, while many other automatons are the equivalent of a lip-syncing robotic boyband.
Beethoven’s 10th symphony “completed” nearly 200 years later
Beethoven’s death in 1827 left behind early notes on his 10th symphony, a work-in-progress. Venturing a guess at his vision—and trying to accurately complete the work—is a daunting task. Nonetheless, a team of musical scholars, composers, and computer scientists attempted it in 2019. They completed their work in time to have it performed for his 250th birthday celebration in his hometown of Bonn, Germany in 2021.
Writing for Smithsonian Magazine in 2021, Dr. Ahmed Elgammal, who headed up the AI portion of this massive task, says that all Beethoven left behind of this symphony was a handful of jotted down ideas and notes. What this team essentially had to do was feed these partially formed ideas into an AI, along with data from Beethoven’s other symphonies and even his creative process, to allow it to become a virtual stand-in for the long-past musical genius. It’s a fascinating look at how computer automation and AI training takes place, where human users don’t so much program a computer, but “coach” it to produce the desired result.
Reactions to the work seem to have been largely positive, with one blogger who attended its performance noting that, “It was incredible!.. I would have never guessed in a million years it wasn’t Beethoven.” On the other hand, researchers only had enough source material to produce two movements, and it would perhaps have been unfair to Beethoven’s memory to extrapolate beyond a certain point.
AI improvisational music created with real instruments
Perhaps it comes as no surprise that with sufficient training, computing power, and time, an AI can produce music. Listening to other players in real-time, while improvising an accompanying rhythm, would seem to be much more challenging.
The Georgia Tech Center for Music Technology and its founder Gil Weinberg have been instrumental in bringing this type of robot to life. Their 2005 robotic drummer, Haile, was able to not only imitate human drummers, but improvise on beats to create something all new. This was followed up in 2009 by the much more capable Shimon, which plays a xylophone using four arms, each bearing two mallets. By 2017 this robot was trained on 5,000 popular songs and two million musical riffs.
Music automation into the future
Will we still be listening to human artists in a decade? In five decades? One would hope so, though perhaps we’ll see AI creep into the more ambient musical genres, such as video backgrounds, hold music and more. In fact, this is already happening, as evidenced by the Luxembourg-based AI music startup AIVA. Right now, you can get access to background tracks that they’ve generated, for a monthly fee.
In his Smithsonian article, Dr. Elgammal makes a comment that, “Most A.I. available at the time [2019] couldn’t continue an uncompleted piece of music beyond a few additional seconds.” Given that he’s talking about just three years ago, it’s an interesting allusion to just how quickly this world is changing.
Perhaps the ultimate goal isn’t replacing human musicians at all, but instead fusing our talents into a human-robot collaboration, producing sounds and effects that would have been impossible before, and where it’s nearly impossible to tell where humanity ends and the robots begin.
See related product
