Tiffany Calvert is a painter who lives and works in Louisville, Kentucky, where she is an Associate Professor of Art and Graduate Director at the Hite Institute of Art & Design at the University of Louisville. She has had an extensive career, including showing her work at Lawrimore Project in Seattle, E.TAY Gallery in New York, the Speed Museum in Louisville, Susquehanna Art Museum in Pennsylvania, and Cadogan Contemporary in London. Her practice combines traditional oil painting and techniques from the Dutch Golden Age with artificial intelligence and machine learning. Here, in conversation with Majuscule board member and artist Amy Wilson, she discusses her process.
Can you contextualize your practice alongside other AI-informed works such as Portrait of Edmond de Belamy, or the “collaborations” between AICAN and Ahmed Elgammal? There’s differences I’ve noted but I’d be curious to hear how you describe it.
I see interesting parallels between the high auction price at Christies for the Belamy piece, and tulipomania, the 17th century Dutch craze for rare and overvalued tulip bulbs. There is an inflated value attached to each that is based on its novelty. Interestingly, tulip growers were most obsessed with a tulip called the Semper Augustus, whose characteristic scarlet striping on white was the result of infection from Tulip Breaking Virus (a blight the tulip growers still must protect their bulbs from, using artificial intelligence). I play off this idea in my own paintings; that tulip growers use AI to avoid mutations, and I create mutations using it. As well as that the word virus is now common to us in terms of illness, mutating viruses, computer viruses.
AICAN (AI Creative Adversarial Network) claims to be the first machine generated artist to pass the Turing test. It uses the same machine learning framework I use (called a GAN, or Generative Adversarial Network) using 80,000 images of artworks. A GAN “learns” how to make a picture of something, usually a face, by studying thousands to millions of sample pictures. With enough examples, these programs become very good at making forgeries. For example, they can make very convincing pictures of people who don’t exist. However, I don’t give the software millions of examples. My dataset, which is composed of Dutch and Flemish 17th c still life paintings, is relatively small – just over 1,000 images. So the AI just doesn’t have enough data to learn from to make a convincing picture. It makes a bad one instead – a mutant version of a spherical flower, for example, wrongly merged with a lemon.
This means that I must react to what the machine learning interprets from my dataset. But do I think the AI processing constitutes artistic decision-making? I think no. I think AI does not pass the Turing test – but I find it a more interesting question to ponder without coming down on an answer. I see a lot of connection between these questions about whether AI can create art, and the art world’s obsession (again) with Hilma af Klint and art made about spirituality and ghosts. We’re very interested again in where abstraction comes from, and struggle to describe by what “magic” artistic decisions come to artists.
Would you describe these works as being, at least in part, about mortality? I can’t help but think of the irony of a computer viewing a vanitas painting – the AI that conceivably could live forever mulling over this depiction of a still life that has long since died.
Yes, those Dutch and Flemish paintings were memento mori; lessons in morality and the fleeting condition of life. In the Dutch flower paintings specifically, not only is a bloom incredibly fleeting (and the painting of it is not), but most of the flowers depicted in a single painting would not all have blossomed at the same time. The question of computers pondering mortality makes me think about corporeality. Even if we finally describe the conditions by which a computer is thinking, and making artistic decisions, it’s still not alive by most peoples’ definitions – it can’t die.
I think these paintings make you think about mortality in no small part because the AI-generated images look like cells or organs. They vaguely resemble those vanitas still life paintings of the 17th century (a voluminous object on a table in a dark setting) but they are a different subject. Not flowers or fruit or dead game, but something both more human and alien. Worse than an uncanny valley – a canyon.
It’s funny you ask this question: I used to say there are only two kinds of artists; artists who make work about death, and artists who make work about sex. I have always been the former but wish I was the latter.
Could you take us through that part of the process? After generating this image, you print it out, and then you paint over it. Why? Why this extra step?
Yes; I have the image printed at large scale, then I apply a mask which preserves parts of the image, and I paint into it. The mask is what creates those hard-edged lines where the paint meets the reproduction.
To be specific, I first collect from internet searches a folder of images of still life paintings. I give the AI my dataset and then I can choose for how long I want it to learn (the longer the more accurate), and how many different images I want it to spit out. In a few hours, it creates the images. I can ask it to produce hundreds of images, and each will be different.
Right now though, this software is very new and it has limitations. It will only give me a small 1000 pixel square image. I then use different AI software to enlarge it many times greater. That AI does a really interesting job of interpreting textures and luminance based on a low resolution image.
I have it printed out with latex based inks on canvas, and stretch it to a traditional canvas stretcher. I apply a full size commercially cut vinyl mask to show and hide parts of the canvas. Then I paint all over it with oil paint – matching the colors and trying to abstractly mimic the textures and forms. Then I peel off the mask. My hope is that in the end, the pictorial abstraction in the reproduction and the painterly abstraction are in some places indistinguishable.
To learn more about Tiffany Calvert and her practice, visit her website at http://www.tiffanycalvert.com.