Duels with artificial intelligence

Exploring relations between painting and artificial intelligence 


y work is often about the relationship between art and technology. I often take as a subject the power of machines to create images. So far I confronted painting with graphic design, then with CGI, then with VR, and then recently with video-gaming. Now I explore the possible relations between my painting process and artificial intelligence 

Above, from left:  Privacy Intrusion Original, 2020, oil on paper, 50 x 70 cm, Privacy Intrusion AI-Version, 2022, oil on paper, 50 x 70 cm.

Here I use artificial intelligence to generate images from verbal descriptions of my existing paintings. I picked a picture that I painted a few years ago, then I instructed an AI algorithm to generate a similar image. The algorithm makes images from verbal prompts. Answering my riddle, the AI image-generator spat out a series of images. None looked much like my painting, but I chose one that matches my original the most, both in composition and narrative. Then I painted the results matching the style of the original. I put both the original and the new painting side-by-side and compared them.

My painting as a riddle

AI's answer to my riddle

Above: Privacy Intrusion Diptych Scheme, 2022, digital graphic

One can describe this project as a conversation with an AI. The goal is to see how AI-generated images match human-generated images in their subtext or their subtlety of meaning. It has the dramatic format of a battle. Other artists can join—if it’s framed as research to find out what is the current level of AI creativity compared to humans. AI vs. Human, like in Kasparov vs. Deep Blue.

Above: Privacy Intrusion,  2022, group of AI-generated digital images, grid view, digital graphic.

It seems like a project where I test the power of an AI algorithm, but in fact it’s about the limits of our creativity. It shows how much human visual creativity depends on media images. One day I’ll lose the battle due to my failure to invent a pure image. Media has infected all my images. It’s obvious that I can’t avoid appropriating images from the media. In fact I fail to create a single image outside the media space. I don’t have any pure untainted images in me. Even the images coming from my subconscious mind I see as shaped through media. Even while dreaming my mind renders—i.e. an ‘archetypical snake’—via media images of reptiles I’ve seen before. I would construct it on canvas using images I remember from the media. Artificial intelligence also uses the vast media archive to train its algorithm. Thus in creating images I don’t have any advantage over AI. We’re sampling the same source. It’s easy to imagine soon an AI trained to generate images on the aggregated imagery of the entire humanity. When that happens my only advantage will be my psychoanalytic randomness. Or something the optimists call a soul.

Above, from left:  Privacy Intrusion, 2020, charcoal on paper, 21 x 30 cm, Privacy Intrusion, 2022, Dall-e2 digital graphic.


Above, from top left:  Privacy Intrusion Original, 2020, oil on paper, 50 x 70 cm, Privacy Intrusion AI-Version, 2022, oil on paper, 50 x 70 cm.