Listen to this story
There’s no doubt that A.I. will play a transformative role in society in the future. In fact, it’s already doing it. In the last five years, artificial intelligence has come from perennial algorithms to become one of the most innovative technologies ever available.
All the big players in the tech industry — Alphabet, Amazon, Apple, Facebook, and Microsoft — are hugely reliant on A.I. and are investing heavily in it for their future ventures. It is yet to be seen which one of them will reach the uncanny valley first.
When we think about A.I., we usually imagine it replacing cashiers at supermarkets or as the voice behind Alexa answering our repetitive requests about the weather or driving Teslas. In recent years, automation has demonstrated the ability to perform technical tasks, such as recognizing faces and places, understanding speech, and writing news articles. But new developments in computer power are making it possible for automation to master skills we once thought were reserved for humans. Like creating art.
In the art field, artificial intelligence has undergone a deep learning process. Fed with thousands of online art databases, the technology now has the ability to recognize a wide range of art styles and movements. This allows it to develop and spew out its own works, ultimately creating new unconventional styles.
But can an algorithm truly imitate human abilities? Is A.I. bringing us into a new era in art?
Some research suggests the answer is yes. In a ground-breaking study, Rutgers University scientists and the A.I. research department at Facebook proposed a system for a generating an art agent called Creative Adversarial Networks. The agent builds on an already existing technique called Generative Adversarial Networks, but adds a component that encourages the algorithm to think creatively.
Introduced in 2014 by researcher Ian Goodfellow, GANs are widely used by machine-learning enthusiasts to create novel, realistic images. The technology compiles a series of algorithms where one network is generating ideas and the other is judging the results. The algorithm loops back and forth until a decent result is reached. As Goodfellow described it to Wired last year:
You can think of this like an artist and an art critic. The generative model wants to fool the art critic — trick the art critic into thinking the images it generates are real.
The Rutgers researchers used GAN as the basis for their own technique but also tasked the judging network with classifying the images into an artistic style. As the generating network tries to make it harder for its critic to classify the art it’s producing, the images become more creative. Since the generator still wants to pass the images off as real, though, it doesn’t go too crazy.
As Rutgers University professor Ahmed Elgammal told New Scientist: “You want to have something really creative and striking — but at the same time not go too far and make something that isn’t aesthetically pleasing.”
Algorithms trained under this technique relied on more than 81,000 paintings from the WikiArt database ranging from the 15th to the 20th century.
In their experiment, researchers also compared the response of human subjects to the A.I.-generated art. The results showed that people could not distinguish between the art created by the algorithm and art generated by contemporary artists. To the researchers’ surprise, the public ranked the automated pieces higher than the images made by humans. This raises the perennial question: Is creativity an exclusively human quality?
People could not distinguish between the art created by the algorithm and art generated by contemporary artists.
It seems humans are already falling for the new art genre. Last month, Nature Morte, a prestigious gallery in Delhi, India, opened its doors to one of the first exhibitions made by the nascent art form. The gallery show, titled Gradient/Descent, features emerging digital artists challenging the dynamic between humans and art. The organizers describe the exhibition as “what art could be in the post-human age.”
When asked about the learning aspects of A.I., Karthik Kalyanaraman, one of the curators of the exhibition, told the Indian news site First Post:
Artists produce art based on the other artwork they have seen. If they didn’t, they simply would not be relevant. Looking at it from a purely pragmatic perspective, if a machine can make humanly surprising, stylistically new kinds of art, I think it is foolish to say well it’s not really creative because it doesn’t have consciousness.
The modern technology is already breaking into the veteran dealing art industry too. The renowned auction house Christie’s recently announced that it will sell the first collection created entirely by A.I. this fall. The A.I.-generated art, developed by the French art collective Obvious, was made using GAN. The main piece, estimated to go for 7,000–10,000 euros and currently displayed at Christie’s London showroom, depicts a French man — a member of the fictional Belamy Family — wearing a black frock coat over a white blouse with most of his facial features blurred. At the frame’s bottom right, one can see the strange signature in cursive Gallic. It reads: 𝒎𝒊𝒏 𝑮 𝒎𝒂𝒙 𝑫 𝔼𝒙 [𝒍𝒐𝒈 𝑫 (𝒙))] + 𝔼𝒛 [𝒍𝒐𝒈(𝟏 − 𝑫(𝑮(𝒛)))]
Big tech companies like Google also know the relevance of art in technology. That’s why they have come up with their own photo algorithm to tackle the last A.I. frontier, creative arts. Developed at Google’s Zurich office in 2014 and released in 2016, Google’s DeepDream is a machine-learning algorithm using a convolutional neural network to enhance pictures and detect faces, patterns, and shapes. It’s the same algorithm that allows Google or Apple to identify photos by their content in our phones.
The resultant pictures often look like pieces taken from a psychedelic-acid trip, or as if a Skynet was on LSD. This is because the network, once given an image, starts a feedback loop trying to recognize what’s in the image from the initial uploaded data sets. The results are endless trippy effects over the same trained “feature.”
So who is the artist of these paintings? The software itself, the programmer, or both?
Professor Elgammal recently told CBS News, “I consider myself the artist in setting up the framework. The machine itself explored the possibilities and gives me answers. So, the machine has a part of the creative process.”
A.I.-made art has the potential to become a tool for artists and art students all over the world. The technology could enhance a person’s talent, acting as a sort of creative collaborator, but one that is not going to ask for a percentage once the deal is done. But can we really consider the resulting artworks to be works of art?
“Working with A.I., artists can harness chaos and complexity to find unexpected signals and beauty in the noise.”
— Rama Allen, creative director of The Mill
What really defines us as intelligent beings is our capability to be subjective, as a form of personal expression. Creativity as well as beauty are abstract concepts that differ with each individual and culture. Throughout history, art has had different meanings and interpretations. A computer that has been coached or programmed into a standardized concept will probably lack the subjectivity and human-like conditions evident in the purest form of art.
Art has always been seen as a pantheon of humanity, a quality that is quintessentially ours that no technology could ever replicate. As we peer into the future, we must remember the great benefits technology has provided us with. Robots will be bounded by the creativity and imagination of the human operating the machine.
A.I. masterpieces are already on track and to deny their future impact would be disingenuous. But until A.I. reaches the desired uncanny valley, humans and robots will have to continue painting, composing, sculpting, and making beautiful, weird, or useless art together.
A human created this piece. No A.I. machines were harmed in the making of this article.