Decoding Machine Art: Music and Painting

Written by Nilesh

April 24, 2017

There are also companies like Juke Deck and Amper Music, among many others, that specialize in creating and training machines to compose music.

Late last year, research scientists from Sony released a song called ‘Daddy’s Car.’ It sounded like just another Beatles song, but the song was noteworthy because it was composed by an Artificial Intelligence machine. The software draws from a massive database of songs, identifies the various small elements that make up a song (the chorus, the guitar riff, the percussion, and so on) and then, like a jigsaw puzzle, brings together these little individual techniques to compose a new song.

There are also companies like Juke Deck and Amper Music, among many others, that specialize in creating and training machines to compose music. These companies mostly seem to focus on using machines to compose background music matching various moods, so you may well have encountered machine-composed music while watching a video online, and not even realized it.

In 2016, Google announced a research project called Magenta, which took up the question of whether AI can produce art—specifically, music and paintings. The project has delivered a piano melody composed by a machine, and has also created many experimental paintings—bizarre, psychedelic images that seem inspired by artists like Salvador Dali. So the question isn’t whether AI-powered machines can compose new music or art on their own.

They already can. The question really is: how well can they do it? For now, the machines are limited to only imitating. That might seem like ‘only’ to us, but for a computer that is incredibly hard to achieve.

What do we truly value in art? It’s rarely the skill, not on its own anyway. When Van Gogh started painting the trees and the night sky in a whole new way, his achievement was not that of physical ability. Any skilled and well-trained painter today can create a painting to exactly match Van Gogh’s style. And these imitations are often so good that the greatest of experts can be fooled. Which is why, in the world of art, there is a small industry at work, busy detecting and exposing forgeries of great artists.

In the same way, any competent group of musicians today could come together and compose a song that perfectly mimics the musical style of the Beatles. When it comes to art, it is not the skill we value so much (that is a given: unless you are skilled, you cannot create art), but originality and perspective. Van Gogh is valued not because of his brushwork, which was certainly outstanding, but because of his perspective, his completely new way to look at the world, that resonates with us even today.

Imitation is in that sense the easy part. Whether you sing an old Beatles song and sound exactly like the Beatles, or whether you create a new song that perfectly imitates the musical style of the Beatles, you are still imitating, and not creating original art. As we saw in our discussion of ‘what is art’ a couple of weeks ago, art is as much a cultural phenomenon, as it is a physical, content creation activity. For us to accept something as art, it needs to be original (not just in terms of content, but also in terms of style), bring new perspective and ideas, and appeal to a new generation of consumers.

In the near future, machines will be able to do it all: create new art in new styles, appealing to new generations of art consumers. Consider music for example. Machines can already imitate well enough. But what if a machine could then learn about how its music is being received? What if a machine could learn about what music is most popular, what type of music appeals to what type of audience, what musical styles seem to be gaining in popularity, and so on? Armed with this type of data, in addition to the database of millions of songs, machines will be able to create and release new music, analyze its reception in specific segments of the population, and keep improving until they become so good that we look to AI machines for new musical styles and techniques.

And this isn’t some far-off futuristic scenario. Much of our music consumption is online: we download music, and listen to music online using services such as Spotify and Saavn. All this data about our consumption activities will eventually power the next wave of intelligence in machines, enabling them to create better, more original, more appealing art.


*This article was originally published on April 24, 2017 in Telangana Today, and can be accessed at Decoding Machine Art: Music and Painting


Related Articles

Peter Thiel’s Blind Spot: What He Gets Wrong in Zero to One

Peter Thiel’s Blind Spot: What He Gets Wrong in Zero to One

Like most of you working in tech and startups, I absolutely loved reading Peter Thiel’s Zero to One. It’s a stunningly thought-provoking book, with a deep new insight pretty much on every page. It’s hard not to be impressed by the clarity of thought as well as the...

Pricing strategies for startups

Pricing strategies for startups

 I’m sure we have all heard of that fraud where tourists walk into a restaurant, order some basic food item-dosa, idli kind of thing-and are presented with a bill for thousands at the end of it. And the swindlers at the restaurant claim that the tourists should have...

Will technology destroy tourism?

Will technology destroy tourism?

What are some things you are aspiring to do this year? Is traveling anywhere on your list? Almost everybody thinks they need to travel. Not convinced? Log in to Facebook, and see what your feed is showing. What friends of yours are vacationing currently? What pictures...


Submit a Comment

Your email address will not be published. Required fields are marked *

Share This