The algorithms are dreaming again.

At Art on Tezos: Berlin this November, pioneering AI artists including Mario Klingemann, Sougwen Chung, Danielle King, Boris Eldagsen, Terence Broad, Mike Tyka, and Von Doyle gathered to showcase work that explores how machines see us, remember us, and sometimes get us completely wrong.

Here, we explore three of these artists specifically: Broad, Tyka, and Doyle, whose work goes beyond technical novelty to ask harder questions about perception, memory, and what happens when we let algorithms mediate our vision.

🫧 NOW LIVE 🫧

»Degazed Body, Disembodied Gaze« by Terence Broad

The AI-generated diptychs, created with @terrybroad's self-developed network bending technique, engage with contemporary debates around surveillance, bias in machine vision, and the politics of representation. pic.twitter.com/aLh9ZCQXQe
— The Second-Guess (@_second_guess) November 4, 2025

Terence Broad: The Hacker Poet

Terence Broad doesn’t just use neural networks, he hacks them.

A senior lecturer at UAL’s Creative Computing Institute with a PhD from Goldsmiths, Broad has developed what he calls “network bending,” a technique that treats AI models not as black boxes to be deployed, but as malleable artistic materials to be manipulated, twisted, and coerced into revealing their hidden logic. His work has earned him the Grand Prize at the ICCV Computer Vision Art Gallery and a permanent place in Geneva’s contemporary art collection, but it’s his latest series that cuts deepest.

Degazed Body, Disembodied Gaze is a collection of AI-generated diptychs that perform a brutal vivisection on machine perception itself. Broad’s method is surgical: he selectively disables the neural pathways responsible for generating specific facial features, then watches as the AI hallucinates what it cannot see. Each diptych presents two versions of the same portrait—one stripped of its eyes, those ancient seats of human connection; another where only the disembodied gaze persists, floating in algorithmic space like a Cheshire Cat’s smile.

The effect is deeply unsettling. When the eyes disappear, the AI doesn’t simply leave a void—it fabricates, confabulates, dreams up something to fill the absence. This is how machines see us: not as whole beings, but as probabilistic assemblages of features, any one of which can be erased, replaced, or reimagined without the system ever acknowledging the violence of the act.

In an age of ubiquitous facial recognition and algorithmic surveillance, Broad’s work asks the most urgent question of our moment: Who controls the gaze when the watcher is a machine?

Mike Tyka – “Rains in Tokyo” (2017). Credit: Courtesy of the artist

Mike Tyka: The Pioneer Returns

Mike Tyka was making art with neural networks before most people knew neural networks existed.

In 2015, he created some of the first large-scale artworks using Iterative DeepDream and co-founded Google’s Artists and Machine Intelligence program. His background traces an unusual path: PhD in biophysics, protein folding research at the University of Washington, then mapping fly and mouse brains at the neuronal level for Google. His sculptures of protein molecules (cast in glass and bronze from exact molecular coordinates) bridge those worlds, making visible the invisible architecture of life.

At Art on Tezos Berlin, Tyka presents work that marks a return. This year is DeepDream’s tenth anniversary, and he showed pieces from 2017: Surrender, previously exhibited in Okinawa, alongside Ascension and Rains in Tokyo, neither shown before.

These pieces avoid the garish, dog-slug hallucinations DeepDream became known for. Tyka’s work with the algorithm is more controlled: a conversation with the machine rather than submission to its fever dreams. His pieces reveal the algorithm as something with its own aesthetic sensibility, its own way of seeing.

ERROR Z sold to @CemDrin 🙌 Thank you! 🌿 pic.twitter.com/9w5t5xgc9t
— Von Doyle (@VonDoyl) November 4, 2025

Von Doyle: The Glitch Mystic

Von Doyle makes classical paintings malfunction.

The Montreal-based artist came to AI art through a circuitous route: fine arts and film production at Concordia, experimental shorts, a video production company that made great content but terrible profit, then a discovery of digital art and the NFT space that felt, in his words, “like a revelation.” His early adoption of Tezos — drawn by low fees, strong community, and what he calls its “punk ethos”— made him a fixture in the Hic et Nunc community.

But it’s his process that sets him apart. Using neural networks, deep learning, and Deep Dream algorithms in Python and PyTorch, Doyle pushes classical paintings through layers of algorithmic processing until they begin to fracture. “Computers dream in a strange, alien language,” he says, and his work functions as translation — or perhaps mistranslation — of that language into something humans can experience.

As part of the show he co-curated for Accompart’s, Doyle presents ERROR A at Art on Tezos Berlin, the opening salvo in a 26-part series running through the alphabet. The piece begins with Rogier van der Weyden’s Portrait of a Lady, a 15th-century masterpiece of Northern Renaissance elegance. In Doyle’s hands, the woman is no longer still. She’s caught in a glitching loop, her reality repeating, fracturing, stuttering endlessly. Her calm elegance slowly dissolves into what Doyle calls “quiet madness”— but crucially, she is not destroyed. She simply cannot escape.

The concept behind ERROR A through Z is built on the idea of a “soft error” in computing: a harmless fault where information flips momentarily, returns slightly changed, but the system continues running. Nothing crashes. Meaning just… shifts. Over 26 iterations across different classical paintings, Doyle traces this shift—faces blur, gestures repeat, backgrounds vibrate, and what was once sacred becomes profoundly unstable.

“This is not destruction,” Doyle insists. “It is quiet malfunction. A memory trying to stay intact, even as it forgets what it once was.”

What They Share

These three artists refuse to treat AI as just a tool. For Broad, Tyka, and Doyle, neural networks are collaborators with their own logic, their own aesthetics, their own strange way of seeing.

Broad breaks the machine to reveal how it constructs us. Tyka converses with it to discover what it finds beautiful. Doyle pushes it until it glitches, finding truth in the error.

Their work arrives when AI-generated imagery is everywhere, often indistinguishable from human-made art, frequently banal. But these artists remind us the interesting question extends beyond whether machines can make art that looks like ours. It asks what we learn about ourselves when we see how machines see us.

In Broad’s diptychs, we confront how machine vision sees exploitable features rather than wholes. In Tyka’s DeepDream pieces, we glimpse an early moment when artists discovered neural networks dream in patterns we barely comprehend. In Doyle’s glitching portraits, cultural memory itself becomes unstable, degrading like a corrupted file trying to remember its meaning.

These works challenge rather than comfort. As algorithms increasingly mediate how we see ourselves (through filters, facial recognition, generative AI, surveillance), art that interrogates machine vision matters. Aesthetically and existentially.

The machines are dreaming. These artists show us what those dreams reveal about who we are, who we think we are, and who the algorithms think we might be. The answer is fractured, multiple, unstable: a soft error in the system that calls itself human identity, glitching forward into an uncertain future.

At Art on Tezos Berlin, that future is already here.