When artifical intelligence destroys repetitive jobs, what’s left for us to do as human beings?
When, in a distant future, that all repetitive work can be done by a machine, where do we turn to?
Although people say that times are accelerating, we may find the same answer as we’d have found 1000 years ago.
At the core of art, a human heart is beating.
Lets explore why.
Consider this famous track “Limit to Your Love” by James Blake.
I remember the feeling I got the first time I listened to it. The subtle silences were filling out an empty space inside of me that no other song had. Somehow Limit to Your Love represented a break from the hectic daily life that allowed me to listen to my own feelings.
Some people may share this experience, others may feel something completely different. What’s important is that I found some meaning in the song and I could feel the artist’s intent of expressing something meaningful. Whether we meant the same thing, I can’t say for sure but perhaps that’s besides the point. It felt new — somewhere a pattern was broken.
Now, when you hear the piece of music inspired by a Van Gogh painting, which Baidu, the Chinese web services company, composed using artifical intelligence it’s easy to rush to the conclusion: Wow, machines are actually capable of making art after all. What’s going to be left for us humans to do?
But before deciding what to think, let’s dig a little deeper.
Baidu did in fact not break any patterns. They recreated them: First they analyzed the colours, objects and setting of Van Gogh’s The Starry Night using machine learning to get an idea of what mood the painting was conveying. Then they created a new song also using machine learning inspired by the pattern of existing songs with that same kind of mood.
There are many layers to this, so lets peel off the first by asking a question.
Were the AI algorithms by themselves actually creating something new?
The first part consisted of extracting the mood, subject and cultural identifiers from the painting. As a person it’s not hard to see how the style, colours and landscape gives the painting a melancholic feel. At this point, however, nothing new is created – we’re merely talking about pattern recognition. The next phase is where it gets interesting.
In the second phase, the system has classified the painting and is now trying to match it with music that has similar characteristics. But instead of using existing music it is creating new tones around that same pattern.
Does that make the machine ‘creative’?
Imagine we change the colour of the painting to red. That would immediately result in a very different and probably more aggressive song. As such, the artificielly intelligent music composer is nothing more than a highly advanced calculator. Change the input and you can be sure to get a certain output.
Yet, we have the perfect illusion of creative machines.
The same principle applies to the artifical Rembrandt painting. I asked Ole Winther, machine learning professor from the Technical University of Denmark, about it and his explanation was quite similar: machine learning in this case is simply a way of recognizing patterns of a certain style and then applying it to something else.
The creators at ING were using machine learning to detect the patterns which constituted the style of Rembrandt; the thickness and direction of his brush strokes, the colours and the motives etc. Then they applied the pattern to different content. Not far from the image filters you’ll find in Photoshop and Snapchat — or perhaps even closer to an art forger: Someone who’s picked up the style of Rembrandt so well that even the best art connoisseurs couldn’t tell the difference. Indeed, the same thing has happened within classical music.
Listen to the song below: Do you think it’s Bach or RoboBach?
Of more than 1,200 people, mostly classical music enthusiasts, half of them couldn’t tell the difference between real Bach and the song above. What does that tell you, when I say that the song wasn’t created by the famous baroch composer but an artificial neural network at Sony’s computer lab?
Again, the same method is at play here: Copy already existing patterns (Bach’s style) to make something which seems like art.
Which is funny if we go back to James Blake, who’s known for saying:
“I’ve always really resented having structure in learning how to express yourself. I’m really more focussed on creativity and coming up with new things. It was too backward for me to study what Bach would have wanted, it’s not one of my concerns really.”
In that quote, James Blake shows himself as the inventive artist pushing for something new. A strong opposite to the examples we’ve seen of using machines to copy past styles of other artists.
Does that, by the way, seem familiar when you listen to the average pop radio station?
Think about each time a new hit song sweeps the world. A few months later, you’ll hear its style echoing through many other pop songs.
Is it perhaps truer that more humans are resembling robots than the other way around?
Whatever Happened to “Steal Like an Artist”?
You may be familiar with the term “Steal like an artist to become one”. But isn’t that exactly what the machines and unoriginal musicians are doing then? Stealing the styles of others?
Again, we have to be careful about attributing intent to a machine. A machine simply follows orders – that is also the case with artificial intelligence as it exists today.
But what about the good artists — don’t they steal also?
To illuminate this, I’m going to steal from the artist, who wrote the book called Steal Like an Artist, to prove a point:
As you can see, there are good and bad ways of stealing.
Thinking about the previous examples of RoboBach, RoboRembrandt and RoboVanGogh, do you think they imitate or transform the original source? I’ll let you decide. What’s not up to discussion, though, is the fact that the Robo-artists only stole from one artist, not many, which I believe leads to the main point:
We have just seen some really crappy examples of robo-art. RoboRembrandt is a poser!
But that’s not the whole truth. In fact, it’s the creators who chose to use machine learning to plagiarize Bach and Rembrandt who are the fake artists.
Machine learning is simply the newest addition of tools in the artist’s arsenal. It’s the brush, violin and synthesizer of the 21st century, all in one.
In this sense, the engineers at Baidu, Sony and ING are not artists. They are a modern version of very capable violin builders and their efforts at plagiarising Rembrandt, Bach and Van Gogh are nothing more than the first attempts of tuning their instrument.
It’s easy to laugh at their attempts, but we will soon see people with true artistic ambitions create extraordinary things with the help of artificial intelligence.
That’s why it’s foolish to assume that you can’t use machine learning to create art simply by judging from what has already been done.
The curious mind will already be exploring the possibilities of AI. The tools are indeed changing at a rapid pace, but the heart of the artist is and will remain human.
Just look at this handwritten note (courtesy of Melbourne’s Arts Centre) from musician Nick Cave listing some of his influences:
Who knows what all these artists have in common and how Nick Cave turned them and God knows what else into this?
A question to ponder for tomorrow: What happens when artificial intelligence gets more sophisticated — will it then create the next Nick Cave? How is art both about extreme complexity and then breaking all the rules and will artificial intelligence be able to do so in the future?
Subscribe below and please share in comments what you loved, liked or hated about this post — It’ll only make things more enjoyable for you to read in the future 😉