(Photo: flickr.com/Dylan O'Dowd)
Writers's block? Just click ‘deep-learning’ for an effortless creative-fix.
What’s the greatest contribution that machines have made to musical composition? Well the most obvious first thoughts would revolve around EDM and “domo-arigoto-Mr.Roboto”. But such were the limitations of machine-generated sounds. Music, as was the case for the entire realm of the arts, was a purely human novelty given that no matter how advanced our machines got, they were but second-rate sidekicks that only provided the means to a creative masterpiece. Only humans had the capacity to conjure from scratch an emotive musical or graphical creation while the stupid robots had to settle for copy-pasting and auto-tuning. I stress the words “had” and “was”.
On Thursday, Google’s Magenta, a project team formed with the specific purpose of programming artificial intelligence to create original music and art, released a 90-second piano melody. The track features a catchy drum-line, hook, bridge and drop, which was all generated from scratch by Magenta's deep-learning-capable AI. Hence, for the first time in history, a machine is credited as an original song's composer.
The Google Magenta team shows world how an AI-composer "doodles"/Image: 9to5Google
The Magenta project was first announced on Moogfest, a techno-optimistic music festival held in Durham North Carolina, and was built on top of Google’s TensorFlow. TensorFlow is a fully open source software library that the Google Brain team has released publicly with the hopes of engendering faster AI development and deep-learning advancements. Very Silicon Valley, anti-Wall St. thinking stuff (not to mention anti-Will Smith #iRobot).
So according to its official statement, Magenta has the specific goal of realizing Google’s ambitions toward machine-generated art. As was the case for the Deep Dream algorithm, which experimented on applying neural networks to images in an attempt to visualize deep-learning and neural networks(the result, a window into the world of schizophrenia), Google has attempted to apply an AI’s unparalleled ability to learn from data into the last vestige of human-exclusive excellence. Creative art. Since AI models have worked so well with speech recognition, translation, self-driving and kicking butt in a game of Go, why not allow it to become not just a tool, but creator of art and music? Oh by the way, those seizure-inducing paintings made by the Deep Dream algorithm sold for $97,600 in an art exhibit last February.
This is how an AI's neural network sees an image, courtesy of Deep Dream/Image: PCmag
Google Brain, the mammoth company's proactive AI-development division, envisioned using neural networks and deep-learning to allow an AI to remember and learn from past musical data. This data could then be used to produce entirely new compositions that are capable of exciting listeners the same way any human-producer could. Hence project Magenta was born. Naturally, the mechanics behind Magenta are highly complex because “music has an element of time”, in which a program must simultaneously link the rhythm with notes and an ongoing melody. Any idiot can keep track of tone and tempo on GarageBand, but these planes don’t come intuitively for a self-operating AI. According to Magenta programmer Douglas Eck, the challenge wasn’t getting the machine to create a tune, but to make it surprising and or compelling; “So much machine-generated music and art is good in small chunks, but lacks any sort of long-term narrative arc.”
As Billboard reports, the music itself was primed with only four notes to start with, but took off from there to create a verse and a bridge. The drum parts were placed as after-effects in order to add zest to what sounds like background tunes to a new Sims expansion pack. The soundtrack is very rudimentary. It probably warrants a healthy dose of ‘find another job’ comments were it posted on an aspiring musician’s Facebook page. But it was only a little more than a week ago when Magenta made its first public debut. As you can see in the video below, Magenta seemingly “doodles” its way into original composition. But as proven time and time again, an AI’s power is in its ability to learn. From rough and seemingly random piano notes, Magenta has given us a workable original piece in a matter of days. Its exponential growth surpasses the pace of any music 101 student in a local university.
Magenta's first-ever baby steps/Video: YouTube
The things that AIs can do never seizes to simultaneously amaze us and scare us. It’s been quite a week for the Google team thus far, as it also unveiled yesterday that its self-driving cars have learnt to honk at pedestrians and other human drivers depending on the level of potential danger it foresees. Pretty soon, it may learn to flip off an errant lane-cutter. However, Magenta is an exciting yet troubling beast of its own. Humans have ceded many titles to its circuited/programmed counterparts. But art and creativity seemed like the last sanctified grounds upon which we humans held a monopoly.
Though it seems unlikely to happen anytime soon, we may in fact have AIs releasing platinum records and directing artists in feature-length albums. Who knows, we may even have an AI-recognition award in the Grammys. With its sleepless diligence and otherworldly analytical abilities, Magenta may be the pioneer of machine-created music labels. Hey, since these mechanized producers come at us minus the disproportionate ego and self-absorbed Instagram profiles, we should consider welcoming this potential game-changer with open minds. They sure as hell won’t sue any poor torrenters or feel the need to tell the world about their penis sizes. Figuratively speaking of course. AI 1. Kanye 0.