For decades now, humans have been using technology to assist with their art, mainly through digital enhancements and editing. That is, making changes within their control. But what about when machines do more of the heavy lifting through artificial neural networks, producing AI art masterpieces on their own with only a couple of cues – or none?
While AI in the creative world is also not exactly new, it represents the next step in machines becoming artists. Artificial neural networks used in AI can mimic human thinking and also learn. An early neural network breakthrough was Google’s DeepDream, which learned how to identify certain objects and shapes after analyzing millions of images. The resulting images gained their own genre title: inceptionism.
Google’s blog on the topic explains it feeds a neural network millions of “training examples.” There are 10 to 30 stacked layers of “artificial neurons,” and each layer talks to the next until the output layer is produced, it notes. Interestingly, the blog also explains that while proven mathematical methods are used to produce results, “we actually understand surprisingly little of why certain models work and others don’t.”
AI Art Generators Becoming More Sophisticated
While many AI art generators require base images to work from, some are promising to create images just by describing them in text. DALL-E from OpenAI is a neural network that says it will create whatever you can dream of just by inputting some words. (I have joined the OpenAI API waitlist, and hope to be able to experiment with it.)
I tried producing some AI art based on my own photographs through Deep Dream Generator, where you can also create a profile and interact with the community. It has predetermined styles (such as a “Starry Night” Van Gogh theme) as well as different levels of inception to choose from. I was surprised by some of the results, especially when you go “deeper” – but some of them were quite pleasant surprises.
Visual art is just one way AI art can take form. The same OpenAI, founded by tech giant Elon Musk, also created a system that can generate text including writing reviews and articles. But perhaps even more impressive is that it can also write poetry, which is thought of as an art that requires a human soul to emote. Meanwhile, Verse by Verse from Google AI is another experimental AI poetry generator that lets you choose your favorite historical poets as muses. Here’s one I produced in the style of Edgar Allan Poe. (I prompted the AI with my own made-up first line. It helped me “predict” each line after.)
springtime is when death departs the sleeping soil
clear into birth, unto a single hour:
Earth, in the shadow of that hall of power
Explored the very blackness of that hour,
Until the darkness was thrown into the star.
I reached within the spirit of the night:
Again I breathed upon my soul was light.
(As a side note, I also originally experimented with a couple of AI writing platforms to produce this article. In this end, I found it easier to write from scratch than it was to edit an AI generated piece to my liking.)
Is There Money in AI Art?
The short answer: yes. In fact, 29 paintings produced by Google’s AI went to auction in 2016, with one of the pieces going for $8,000. An AI art piece, Portrait of Edmond Belamy, later sold for $432,500 at Christie’s. This latter piece was produced using an algorithm by a Paris collective called Obvious that uses a method it calls GAN (generative adversarial network.) The system was fed thousands of portraits from as early as the 14th century to produce the result.
Platforms like Deep Dream Generator are offering use of their service for a monthly fee (shout out to the team for the extra credits to experiment for this article.) Meanwhile, there are already several artists from across various disciplines making a name for themselves through their AI art, such as Sougwen Chung who collaborates with AI-guided robots.
But will AI art neural networks take away money-making opportunities from their (for now) human overlords? It’s possible that entire fictional novels could be written, complete with complex characters that think and feel in believable ways. (Apparently, some AI novels have already been produced such as Dinner Depression that also features an AI art cover.) Perhaps in the future, AI musicians will find each other online using complex algorithms for compatibility, and then go on virtual tour. It may not be so far-fetched, when you consider AI can generate music – highlighted by a recent project that created “new” songs from late artists to shine a light on mental health.
Will Artificial Neural Networks Overtake Human Creations?
Some say no to this question, predicting AI will continue to be a tool, with the artist still remaining in the creator’s seat. But what of an artist’s soul, which arguably makes them human and unique? Can a machine mimic this? Not surprisingly, the subject of whether AI could have a “soul” has come up in the past. A BBC article explores this topic – and while it doesn’t dismiss it, it notes what is considered the “soul” could be explained by science. Therefore, perhaps a machine doesn’t need a soul at all to accurately portray a human – and create original works of art.
For me, the question moving forward will be: do we want our art to be automated? Do we want to give up the human experience of making art and expressing through art? While an algorithm may faithfully reproduce a work of art, will it ever replace the required experience it took to reach the concept in the first place?
Or has AI art already achieved that?
