Ivona Tau on collaborating with AI’s beautiful unpredictability. ⛯
In a Gemma Q&A, the artist Ivona Tau shares how she uses neural networks to create new and unusual images with a surreal touch.
Ivona Tau works with neural networks as a medium—drawing emotion out of artificially intelligent tools. She has often used her own photographs captured on analog and digital film to train generative neural networks (GAN), in order to create relatable memories.
In a conversation with Gemma co-founder Eileen Isagon Skyers, Tau talks about her hopes and speculations for AI technology. She also describes the creative process behind her Founding Artist Edition, “Surrealist's Dream: /Imagine Femme Fatale by Dora Maar.”
Tell us about your new piece, titled, Surrealist's Dream: /Imagine Femme Fatale by Dora Maar. How did you make it, and how did AI play a role?
I’ve always been inspired by surrealist photography and its efforts to tap into the creative powers of the unconscious. AI models are a natural extension of the early 20th century experiments with montage, solarization, and experimental photography.
A few years ago, I started gathering a dataset of surrealist photography images from artists such as Man Ray, Hans Bellmer, and Lee Miller, among others. This collection encompassed a wide variety of themes and visual styles. Using AI, I deconstructed the surrealist aesthetic, which is already abstract. I trained a custom AI model on these images to capture the movement's essence without mimicking any particular artist. The results were fascinating—few realistic elements, and each with their own unique logic.
How does this piece connect to, or diverge from, your larger practice?
This project was quite different from my usual work, and I was reluctant to share it for a while, keeping it in the drawer of "works in progress." I often only work in color, and the model trained on old photographs was obviously monochrome. For Gemma, I decided it was a perfect time to revisit the “Surrealist’s Dream.” The AI toolset has developed so much in the last few years, especially with the introduction of high-res text-to-image generators. It made returning to the project feel like an entirely different world.
When I began revisiting this process, I noticed something peculiar. When you search Google for old photographs, the results are often images generated by Midjourney and DALL-E. It’s already difficult to filter between actual archival images and AI-generated ones. I found this quite ironic. I decided to embrace the AI-generated imagery by including it in the extended training dataset, adding yet another layer of abstraction.
What attracts you to collaborating with AI? What aspects do you find exciting, worrisome, or frustrating?
AI has a very distinct viewpoint compared to a human. It can generalize large swaths of information and process details differently than you or I would. That’s why I love working with AI: it allows me to extend my gaze and expand my perspective. In the process of training custom GAN models, I don’t use any text or direct prompts. It can be frustrating to see the model produce something unexpected, but that’s part of the beauty of this unpredictable medium.
What do you envision as the future possibilities for AI? What do you want to explore as the technology continues to become even more advanced?
I'm curious about what comes next after the text-to-image phase in AI. Since a substantial part of our communication and intention is conveyed through non-verbal cues like gestures, there's a lot of opportunity for further exploration.
“I’m not interested in recreating my visual language through AI. I just want to pursue a sense of discovery and surprise.”
Read our recent interview with Anna Condo.
Gemma is an emerging community co-inhabited by artists, curators, and contributors with a shared vision for how culture can shape our future. This summer Gemma is releasing weekly open editions from our Founding Artists, which you can mint now. Subscribe and follow us on Instagram and Twitter to be notified about new releases.