When machines collide with the world of art.
If ancient philosophers caught wind of the artificial intelligence humans would go later create, they’d no doubt wax poetic about the selcouth beauty in the idea of those creations sharing in what we believed to be an ability limited to only humans: art.
Art is the fabric of human expression.
The earliest humans used paintings and drawings as a medium of facilitating communication; but with humans once again pushing the limitations and boundaries of technology, discoveries become unearthed and advancements are made.
This of course imposes critical discourse (however inadvertently) on us to demarcate the attributes of what makes us human and what makes something robot.
This deceptively straightforward task is more demanding of considerable thinking and analysis than it lets on. The more we unearth, discover, and charter new territory in this unknown, the murkier the answers we believed to be clear become.
Art’s nature is inherently subjective.
Where someone may see a flower, another might see the beginnings of a curved elephant trunk or a piece may be met with glowing enthusiasm by a legion of art zealots but simultaneously face sharp criticism from another connoisseur.
Philosophers have long debated how to succinctly condense art into a definition but were largely met with unsuccess. The difficulty in summarizing what constitutes and what doesn’t as art however creates a capacity of negotiation for AI art — there’s no certain criteria for something undefined.
However for the most part, there is a social consensus that exists that art should fall within the realm of being i) aesthetically pleasing ii) emotionally expressive.
Any artist will tell you that art is a process. Years and years of deliberate practice compounded result in artistic ability; the skill does not reveal itself within a night’s work.
Artificial intelligence also has the potential and capacity to learn how to create art and they’re able to through Generative Adversarial Networks (or GANs for short).
The GAN system learns through training data of 81,500 examples of paintings which creates the foundation for knowledge of different artistic styles.
Generative Adversarial Networks are composed of two AI neural networks: a generator and discriminator.
Diagram of an Artificial Neural Network (ANN) Image courtesy of ResearchGate
The generator produces images for the discriminator to then analyze the image passed and decide whether to classify the image as output from the generator or a painting derived from the training set.
Data scientists then feed noise into the generator and from there it relies upon the trained weights to produce the output.
The discriminator trains through analyzing pre existing paintings created by humans (expected output is 1) and the art produced by the generator (expected output is 0).
The rivalrous nature of the GAN system is designed to adjust and improve both models simultaneously — the generator adjusts its weights to improve performance based upon feedback provided by the discriminator meanwhile the latter becomes better at discerning fakes from reals.
Diagram of a Generative Adversarial Network (GAN) Image Courtesy of GeeksForGeeks.
To train the model, the conventional process of art production with AI was altered: images that could be classified into preexisting paintings were rejected from use in the data.
A move feasible through “maximizing deviation from established styles and minimizing deviation from art distribution” or to say that images that violated the terms (having been a preexistent painting) were intentionally curtailed.
The inherent beauty in Art lies beyond the outer layers of visual pieces though; art isn’t limited to a certain source. It’s an experience that transcends aesthetics. The scope of AI’s artistic ability is also extended to:
Should she kneel be?
In shall not weep received; unleased me
And unrespective greeting than dwell in, thee,
look’d on me, son in heavenly properly.
Plays tell stories — they unpack things of social importance and transform them into a general viewing experience. AI also has the capability to do this. Built with a Recurrent Neural Network and TensorFlow, playwrights may no longer be synonymous with being only human.
- The creation of poetry
“the sun rays struck my face warm tingles to my fingertips the light showed me a path should walk down i spoke and the whispers of the breeze told me to close my eyes i lost my way in a paradise”
How do you train something unfeeling as a robot to pass off poetry? Something that even humans with all their sentiments and life experiences find notoriously difficult to translate feelings into words? Microsoft and Kyoto researchers were faced with this query as they embarked on a philosophically complex exploration to bridge word art and AI.
Fed on image data with a combination of human provided descriptions/poems, the results were fruitful: when judges were presented with the poems created by Artificial Intelligence, discerning what was human and what was robot was not an undemanding task.
- The composition of music
Trained on cosmic amounts of audio data, AI can learn musical theory and produce musical compositions. You can now pay for services that provide you with access to these sounds.
It isn’t too far of a cry to associate AI with art. In fact, robot art is subjectively (there’s that word again) more effective in the pursuit of beauty, a named main principle of art, than work created by humans.
However, the belief that AI art is a weak imitation of the emotional expressions born into the world through humans may just be a direct result of our own biases. The presence of this thinly veiled prejudice was revealed in a social experiment where scientists took samples of human art and works by AI; unbeknownst to the test subjects.
Most peculiarly, the AI art was declared by the human judges to be of higher novelty, complexity and inspiration.
I don’t foresee including many of my own personal opinions in my future articles, but it would leave me with a phantom itch if my thoughts weren’t fully expressed on this topic. When I began learning about AI a couple of weeks prior, the technology was exciting but the full scope of what this technology’s potential wasn’t revealed.
I believed AI to be mechanical and extremely capable; the capability extending beyond the reach of humans, but still limited.
They may be able to diagnose diseases better than trained doctors and give better editing suggestions than the typical writer, but lack the condition of being human, or so I thought. The beauty in sharing knowledge is that you also happen to learn.
Throughout the process of writing this article, the full magnitude of what exactly AI can do finally hit albeit delayed.
There is something so strange and magical yet structured and logical about this technology. AI runs on complex math, code, and data, but doesn’t simply wade in the shallows with non-esoteric tasks.
The nature of this technology is designed to make us ruminate further about what makes us human, and that lies beyond skin and bones, code and data.