Take a short break and listen to the following piece of music. Sounds catchy?

The tune is simple, but pleasant, made with the general positive preference of the human ear in mind. That specific musical piece, my friend, is one of the masterpieces of Google’s Magenta program. It is music, composed, and performed entirely by artificial intelligence.

Deep learning neural networks have allowed AI to break into many spheres of professional work over the decades of the technology’s improvement. But none is perhaps more important than its actual foray into the one thing that we still think makes humans unique: creativity.

Original ideas into unique algorithms

Indeed, art, music and writing are rather abstract, expressive, and subjective. It is often related to the culture of whoever is struck with that creative genius to build, or fabricate, whatever it is that drives his or her inspiration.

But what is creativity? Oxford dictionary gives us this rather straightforward definition:

The use of imagination or original ideas to create something.

The intent to form new concepts and convey them with a unique purpose has so far been an elusive element to AI. This is due to the inherent limitations of traditional computers and network systems. Processors execute instructions. Software runs programs. But they do not typically generate information on their own using only a limited set of integrated data.

At least that was the case until the advent of artificial neural networks. Contrary to the traditional direct number crunching capabilities of supercomputers, such neural networks are structured in the same manner as a brain neuron. These networks produce the initial data and also provide the necessary associated connections to make sense of things.

In simple terms, it is like a multitude of tiny bots, built to progressively learn from each last set of bots, accumulating more and more data until it forms a complex web of algorithms representing the task it is built optimize.

This is vastly different from direct programming, in that instructions are not constantly provided, but are instead generated, or learned, using a base set of initial data by the artificial neural network.

So how does this relate to creativity? Remember, the official definition of creativity is to “use original ideas to create something”. Artificial neural networks generates its own instructions, and optimizes it as it continues to progress at each task cycle. This is quite similar to an artist improving upon a style or skill set through the years, with the critical difference being the AI can receive several lifetimes’ worth of information in just a brief span of time.

What it sees is what it does

Now, back to Magenta. The Google Brain team designed Magenta to “learn” music using computational extractions of all existing musical works. That is, Magenta crammed the entire music library of the human race, and studied every bit of data associated with it. It then developed its own algorithms to help it compose its own music based on what it has “learned”. The result is then what you hear at the start of this article.

This is basically the same template that almost all deep learning systems today use to create their own derivate artistic works.

Torch for example, is a scientific computing framework and machine learning library, with one of its possible features being that it is capable of writing and generating similarly structured scripts based on what it reads. Artistic AI aficionado Robbie Barrat even teaches his AI to study and reproduce classical paintings after studying them, to rather unsettling and sometimes horrifying results.

As we begin to unravel these accomplishments of deep learning AI over the course of several years, we then eventually see an underlying connection between perception and creativity. Perception, in the sense of how the neural network interprets the data, and creativity, by how it generates its own data based on the information it processes.

Michaelangelo was known to have said that:

Every block of stone has a statue inside of it, and it is the task of the sculptor to discover it.

Creativity is - within this given context - seen as the product of the AI’s view of the world, and how it eventually molds this view.

Conclusively, there is indeed the potential for AI to be artists, because what it sees is what it learns, and eventually, what it does.

Human appreciation of non-human art

Ultimately however, regardless of whether the AI could generate images, musical compositions, or written work, the judgement of whether we could actually consider it as art is still up to us humans, at least for now.

Appreciation of art typically includes the meaning, intention, and the effort of the artist that made the work. After all, we understand the effort it took to finish a masterpiece.

We know that there is intent with the basic desire to project one’s inspiration onto a tangible object. Lastly, we see meaning to the finished product due to our personal perceptions, regardless of whether that is actually the original intent of the artist or not.

Because of this, accepting derivative work created by an AI is perceivably much more difficult. It doesn’t seem that the AI had the effort to finish the work. Its intent is basically non-existent because it is not sapient or intelligent enough. Meaning is basically lost, because oftentimes we fail to perceive what is even meant to be depicted in the first place.

One interesting case is the book Harry Potter and the Portrait of what Looked Like a Large Pile of Ash. This rather mouthful title was actually a fan-made project made by Botnik Studios.

Currently consisting of only one chapter, it was made by letting a text-predicting AI study seven original Harry Potter novels. Take a look at these two lines from that one single chapter:

The castle grounds snarled with a wave of magically magnified wind. The sky outside was a great black ceiling, which was full of blood.

Putting certain word absurdities aside, the sentence does actually have a semblance of sense. At the very least, it was oddly capable of maintaining a consistency resembling that of an actual Harry Potter novel.

But do we even consider it as a fanfic? Most likely not. It is fun to check out though, and is quite a recommended read for those who want some nice random laughs.

Then again, we must also be wary of the reverse: giving special, separate treatment to works of art simply because it was made by an AI. To give it special criteria is in the end, still treating it not as a work equal to that of human artists, but just an object of fascination.

If that is the case, then AI derivative works are still far from being comparable to that of true human-inspired art, music and literature.

Non-human appreciation of human art

In the end, we may be holding back the development and appreciation of AI art because deep inside, we still see the strings of human control at work. Indeed, the initial data it has to learn should still come from a human user. It won’t independently create something out of the blue just because it had its own electronic stroke of genius.

To quote Mark Reidl, associate professor at Georgia Institute of Technology, in an article from MIT:

Neural networks are kind of in the imitation mode. You can pipe in the works of the classics and they’ll learn patterns, but they need to learn creative intent somewhere.

So can AI today replace artists, writers and musicians? Not just yet, but soon.

Photo by Yannis Papanastasopoulos on Unsplash