[with Eda Özbakay]
Mario Klingemann is regarded as a pioneer in the field of artificial intelligence-based art, neural networks, and machine learning. His work revolves around human perception of art and creativity, and how machines can enhance or emulate these processes. He has collaborated with numerous institutions, including the British Library, Cardiff University, and the New York Public Library, and was an Artist in Residence at Google Arts and Culture. His works have been exhibited at the MoMA and the Metropolitan Museum of Art in New York, the Photographers' Gallery in London, the ZKM Karlsruhe, and the Centre Pompidou in Paris. Klingemann received the British Library Labs Artistic Award in 2016 and, in 2018, won the Lumen Prize Gold Award, which celebrates artworks created with technology. He was also honored with a Mention of Honor at the 2020 Prix Ars Electronica. His installation Memories of Passersby I made history in March 2019 as the first autonomous AI machine to be successfully auctioned by Sotheby's. His most recent projects include Botto, a decentralized AI artist, and A.I.C.C.A., a robotic dog that serves as an art critic, whose debut exhibition took place in June 2023 at Espacio SOLO (Madrid).
When did you first become interested in the potential of AI as a creative tool?
I belong to the first generation that grew up with home computers, immersed in the process of digitization that began in the late 1970s. At 12, I received my first programmable calculator, followed by my first home computer. Since then, I’ve been fascinated by creating through code and working with digital images.
My first encounter with artificial intelligence dates back to the late 1990s, when I came across a book by Marvin Minsky titled Mentopolis in German—though I believe the original title was The Society of Mind. Although it was a purely theoretical text, I was captivated by the idea of a separate intelligence capable of making decisions autonomously or assisting me in my research. In hindsight, his ideas differ from those I rely on now, but at the time, I found the perspective incredibly stimulating. In the 1990s, there were early experiments with small-scale neural networks, but they were still very rudimentary, primarily mathematical in approach, and not applicable to art. Back then, I didn’t have access to machines capable of operating in certain ways; the entire technological development wasn’t yet mature.
My first concrete project in creative automation dates back to 2008, when I developed a generative tool for creating images. The main challenge in generative art is selecting meaningful results from an infinite number of variations. I worked to automate this selection process even before the advent of deep learning, but the idea only came to fruition when deep learning became a reality about ten years ago. I found it fascinating from the start because it was exactly what I had always wanted to do.
Deep learning opened up new possibilities: the image classification models that emerged at the time were still imperfect but extremely promising, far surpassing previous achievements. Later, this process was reversed with the arrival of models like Google’s Deep Dream. Generative Adversarial Networks (GANs) came shortly after, enabling the creation of realistic images. What I find most fundamental in all of this is the concept of "latent spaces"—the internal representations learned by models to understand how the world looks and its connections—because it allows us to explore image or meaning spaces in innovative ways.
