AI art is here to stay

Some paintings grab your attention, stop you in your tracks and create an impact that is at once deep, at times undefinable. The style of great painters in history is so distinct that you can immediately identify whether a painting is the work of Vincent van Gogh, Picasso, Monet, Rembrandt, MF Hussain, SH Raza, Raja Ravi Varma, or VS Gaitonde. Their work is priceless, most of them are in museums and private collections, and if at all some items reach auctions, they go for large sums of money. Work by some artists is sold for sums many times the reserve price, while others struggle to reach the basic amount, and remain unsold.

One auction at Christie’s New York created history of sorts in 2018. A portrait in a gilt frame depicting Edmond de Belamy, a portly gentleman, appeared unfinished and vague, but was sold for a whopping $432,500 (more than Dh1.5 million), shattering the $10,000 estimate — almost 45 times the reserve price. Paintings have long been sold for higher sums, but this one was unique. It was not the product of a human mind: it was the first AI-generated portrait ever sold. At the bottom right, where the artist’s signature is usually placed, was an algorithm, defined by an algebraic formula with many parentheses. The auctioneer and others heralded it as the arrival of ‘AI art’ on the world auction stage.

That painting, “if that is the right term”, as Christie’s puts it, was one of a group of portraits of the fictional Belamy family created by Obvious, a Paris-based collective of researchers, artists and friends, who are working with the latest models of deep learning to explore AI’s creative potential. Their method goes by the acronym GAN, which stands for ‘generative adversarial network’.

Says Hugo Caselles-Dupré of the collective: “The algorithm is composed of two parts. On one side is the Generator, on the other the Discriminator. We fed the system with a data set of 15,000 portraits painted between the 14th century to the 20th. The Generator makes a new image based on the set, then the Discriminator tries to spot the difference between a human-made image and one created by the Generator. The aim is to fool the Discriminator into thinking that the new images are real-life portraits. Then we have a result”. The collective initially worked with landscapes and other genres, and fed the algorithm sets of works by famous painters, but Caselles-Dupré and colleagues found that portraits provided the best way to illustrate their point, that algorithms can emulate creativity.

Does the portrait’s auction suggest that there is a future for AI-generated art?

Says Richard Lloyd, International Head of Christie’s Print Department, “It’s a portrait, after all. It may not have been painted by a man in a powdered wig, but it’s exactly the kind of artwork we’ve been selling for 250 years. AI is just one of several technologies that will have an impact on the art market of the future, although it’s far too early to predict what those changes might be. It’ll be exciting to see how this revolution plays out.”

Type in #AIArt on Twitter or Instagram, or ‘AI artists’ on Google, and you get an idea of AI’s vast, imaginative strides.

Today, many AI artists produce and sell work online, attracting increasing interest and income. ‘Creative AI’ is a field, with experts, researchers, artists, and fans, spanning the range of creativity, including games and music. Used initially to make scientific discoveries based on huge data sets, AI has entrenched itself in everyday life. As it moves from new feeds to productivity tools, the boundary between what is human-made and machine-made is becoming blurred, altering aspects of how we perceive and interact with the world. Creativity was long considered a unique human trait, but dip into the academic and popular discourse of ‘Creative AI’ and machine learning, and you join a raging debate. How does one deal with the issue of 
bias in data sets fed into the machines? How should AI art be curated? How to find a common language beyond algorithms? How and where should artists learn AI skills? Is the algorithm that created the Belamy portrait its author? Says Caselles-Dupré: “If the artist is the one that creates the image, then that would be the machine. If the artist is the one that holds the vision and wants to share the message, then that would be us.”

It is no longer impossible to order your ‘self-portrait’, a landscape you love, or an abstract painting produced in the unique style of van Gogh or Salvador Dali, if you so desire, even if it will not be art but mimicry.

As historian Yuval Noah Harari puts it, “Humans are essentially a collection of biological algorithms shaped by millions of years of evolution”. Several experts across the globe have been working on AI art, including Ahmed Elgammal, director of the Art and Artificial Intelligence Lab at Rutgers University in New Brunswick, New Jersey, where he is working with a system called CAN (based on similar lines as Obvious’ GAN) but stands for ‘creative adversarial network’. The CAN can be specifically programmed to produce novelty, something different from what it sees in the data set, which includes paintings from the 14th century onwards.

Who are AI artists?

AI artists are mostly those who came of age in the age of computing and the internet, but others, including those in south Asia, have begun thinking of incorporating AI in their work. Says Mumbai-based artist Prakash Bal Joshi: “AI has created perhaps the most critical challenge since the development of cameras. This time it is going to change the way artists create work as well as how society deals with the art world itself. It is no simple mechanical advancement of technology but with AI, the machine will analyse, learn, imagine, and create artwork without any involvement of human beings. Artwork created by AI is a craze, being sold in international auctions and finding many buyers. For me, I wonder whether it is going to be a race between me, my creativity and AI-led machine creating artwork. Who will ultimately win? Only time can tell, but I will be avoiding doing or creating anything that the machine has already started making. Though at one level AI challenges me, I may be able to use AI to avoid repetition or use AI for repetitive work or use AI to reach out to the specific art collectors who may like the kind of artwork I do. Ultimately, the ingenious human brain will find an echo system to deal with this new challenge.”

‘Creative AI’

As many people embrace AI to create art and some remain wary of it, a new report released last week by researchers at the University of Oxford makes a significant intervention. Titled AI and the Arts: How Machine Learning is Changing Artistic Work, the interviews-based case study notes the explosion of interest in ‘creative AI’ but concludes definitively that machines eventually will not replace artists. The team comprised researchers from the Oxford Internet Institute (OII) and the Department of Engineering (reflecting the interdisciplinary nature of the subject), and interviewed leading artists, curators, and researchers in the field of ‘creative AI’. It found that artists highlighted a difference in scope between human and machine creativity. While machine learning models could help produce surprising variations of existing images, practitioners felt that the artist remained irreplaceable in giving images artistic context and intention.

Robbie Barrat, one of the artists interviewed, had a different view of the Belamy portrait auction: “The narrative that GANs are creative or whatever is gaining in popularity. It’s a strong thing to market. We saw this at Christie’s, which auctioned off that Obvious piece, and the whole narrative attached to it was ‘a robot made this’ or ‘a computer made this’. They’re not looking for a dialogue between a fashion designer and an artist that works with algorithms. They’re looking for a narrative of ‘computers can design clothes on their own’. I don’t know why you would want that. It’s a lot less interesting than the real narrative of creatives using AI or machine learning to produce their works. I don’t know why people are so attached to that first narrative and want it so badly.”

The study’s main message, says OII researcher Anne Ploin, is that human agency in the creative process will never go away. She says: “Parts of the creative process can be automated in interesting ways using AI (generating many versions of an image, for example), but the creative decision-making which results in artworks cannot be replicated by current AI technology. Artistic creativity is about making choices (what material to use, what to draw/paint/create, what message to carry across to an audience) and develops in the context in which an artist works. Art can be a response to a political context, to an artist’s background, to the world we inhabit. This cannot be replicated using machine learning, which is just a data-driven tool. You cannot — for now — transfer life experience into data. AI models can extrapolate in unexpected ways, drawing attention to an entirely unrecognised factor in a certain style of painting (from having been trained on hundreds of artworks). But machine learning models aren’t autonomous. They aren’t going to create new artistic movements on their own – those are PR stories. The real changes that we’re seeing are around the new skills that artists develop to ‘hack’ technical tools, such as machine learning, to make art on their own terms, and around the importance of curation in an increasingly data-driven world…Don’t let it put you off going to art school. We need more artists.”

Those interviewed in the study reflected on a range of academic and practice-based implications, including mystifying AI art for commercial reasons, hype, and public perception whether AI art is really art. Neural artist Sofia Crespo told the researchers: “Art is still going to be around human expression and human emotions. That’s the way that art works. I had a few people tell me that my work isn’t really art. That it’s made by an algorithm, and not by me, and therefore it’s not art. But there’s a real human doing this data stuff. It’s me. I’m a real person feeling something. Why is that less art? In a way, I think having to ask ourselves these kinds of questions means that AI is somehow changing the field.”

The study notes that in the history of art, glitches are often artistically desirable. Art produced by machine learning is no exception to this: while the capabilities of machine learning models were valued by the respondents, most were particularly interested in their edges: the artistic potential of machine failure. As Barrat explained, “I really wanted to introduce some sort of misinterpretation. I thought that the landscapes were a bit boring because the network got it right. We have so many landscapes. It just seemed boring for a neural network to produce more plain old landscapes.”

Helena Sarin, another respondent, told the research team that if machine learning models “got too good”, she would have to find other tools: “GANs are not perfect and that’s why I work with them. I use two types of GANs, one is maybe three years old, and the other over a year old. I intentionally don’t upgrade to new advances because they push the more photorealistic stuff, which is exactly against my process. If they start going photorealistic, I’ll have to find a new branch in this area”.

There is a buzz in the field and the Oxford study and rapid developments in AI art reflect new impulses in a direction that is yet to stabilise as a genre. As artist Jake Elwes told the study: “That’s why this is a fascinating time. Saying ‘this is a new medium’ or ‘art movement’ or ‘genre’–– as some people are saying –– seems very grandiose, but it’s also exciting. The rulebook hasn’t been written yet, and that’s such an exciting place to be. We don’t have the answers to these things. We’re constantly repositioning ourselves and working out our views.”