Can AI Break Through the ‘Cacophony of Pretty Pictures?’: In Conversation With Mario Klingemann

Key Takeaways

• In 2015, Google’s DeepDream inspired artists like Mario Klingemann and helped launch a new artistic movement.
• Today, AI art has caught the attention of major collectors and auction houses.
• Increasingly autonomous AI artists almost completely remove humans from the equation.

In the mid-2010s, Mario Klingemann emerged as one of the first artists to start experimenting with what were then cutting-edge AI tools. But in the decade since Google’s DeepDream kickstarted the modern AI art movement, the means of machine-generated imagery have gone mainstream.

Nonetheless, Klingemann is confident that new techno-artistic projects can continue to push boundaries and break through the “cacophony of pretty pictures” stemming from mass-market AI image generators.

The DeepDream Revolution

Developed for the ImageNet Large-Scale Visual Recognition Challenge , DeepDream turned the field of computer vision upside down in 2014.

Up to that point, developers were preoccupied with training neural networks to recognize images. But when Google engineer Alexander Mordvintsev ran the process in reverse, the idea of using pre-trained models to generate images was born.

In an interview with CCN, Klingemann recalled being struck by DeepDream’s psychedelic outputs, which he identified as a “pivotal moment” in the history of AI art.

The software was “the first time where you got something that was almost out of this world and which was, at first, really incomprehensible and a very different aesthetic,” he observed.

Klingemann’s Autonomous Art

After Google released DeepDream to the public, its idiosyncratic style, full of cats, dogs, not-quite faces, and architectural features, eventually grew boring. But Klingemann and his peers soon took the technology in new directions.

By training computer vision models on custom image libraries, AI artists turned them into unique digital paintbrushes. But what made the emerging art form unique was its capacity for surprise.

Klingemann’s 2019 piece “Memories of Passersby I” perfectly encapsulates this dynamic. Powered by an AI model trained on thousands of portraits from the 17th to 19th centuries, the installation generates a never-ending stream of images, each one unique while still conforming to an overarching style.

For Klingemann, however, adopting the new image generators was just part of a broader artistic gesture.

“The whole idea that I could automate parts of my creative process has been with me since I started using computers.”

Botto, the AI Artist

For his latest project, Klingemann took the concept of automation a step further.

Designed in collaboration with the software collective ElevenYellow, Botto is a kind of autonomous AI artist that generates its outputs without any human intervention.

Botto’s “art engine” generates its own prompts using ChatGPT-style language models. It then feeds these into different text-to-image models to create a series of unique artworks.

Even the process of curating the best outputs out of about 20,000 images produced each week has been automated.

First, an AI “taste model” selects 350 outputs from the larger pool. From these, a decentralized autonomous organization known as BottoDAO votes on which images should be selected to become canonical artworks.

Completing the loop, those votes are fed back to train the prompt generator and taste model for the next week.

“Initially, Botto started as an artwork of mine,” Klingemann observed. But after a certain point, he had to “cut the umbilical cord” and let the system grow on its own.

“It’s about letting go and trusting the system that you build to continue with the values you passed on to it.”

Since relinquishing control of the project, he has questioned the DAO’s choices at times, but ultimately, “that’s what makes it interesting,” he stressed.

Machine Artists Develop a Life of Their Own

Botto is part of a new generation of AI machines that are growing a life of their own independently of the artists, technologists and other collaborators that stand behind them.

One of the most noteworthy examples of this trend is AI-Da, a robot artist whose piece “AI God” recently sold at Sotheby’s for $1 million.

Over the years, artists like Harold Cohen, Jean Tinguely, and Chris Chen have explored notions of ownership and creativity by building machines with the capacity to surprise.

Platforms like AI-Da and Botto build on this tradition, deploying the latest advances in AI and robotics to interrogate what counts as art and what counts as an artist.

To Simon Hudson, the ElevenYellow programmer who collaborated with Klingemann on Botto, such preconceptions are the inevitable starting point for any AI art.

“You start with a model of a human artist, right? It’s very skeuomorphic of what does a human artist do to succeed? And then you realize, okay, well, this is a machine artist. There are things that they can’t do that human artists can,” he noted.

At the same time, AI art platforms from DeepDream onward have evolved to output things that don’t merely emulate human creations.

When Botto shared an image that recalled “Dogs Playing Poker” by Cassius Marcellus Coolidge, the DAO initially jumped on it; Hudson observed:

“After that came a conversation of, well, maybe we should be pushing Botto to do something new, to do something that a human artist hasn’t done.”

AI Style

While the BottoDAO may be pushing artificial intelligence to be more original, other applications of the technology are taking the opposite approach.

Midjourney and Stability AI, the developers behind two of the most popular text-to-image models, have been the target of lawsuits by artists who feel like the platforms are ripping off their signature style.

The idea that something as ambiguous as artistic style could be protected as intellectual property is still a new and untested legal argument. However, with the new technology increasingly viewed as threatening artists’ livelihoods, such lawsuits could set an important precedent.

For his part, Klingemann is against what he called “feudalism in intellectual property space, which he argued expands the strictly delineated borders of copyright into “a huge portion of possibility space.”

Although he sympathized with the desire to protect a recognizable signature style from copycats, he stressed that artists have always been inspired by each other.

“You made it. You shared it with the world. You have to allow others to be inspired by it. And these days, that includes machines.”