TSM - Pioneering AI Synergy: A Conversation with Lena Thorsmæhlum and Thordur Arnason of Gervi Labs

Ovidiu Mățan - Founder @ Today Software Magazine


Interview conducted with assistance from synthetic lab member AIson using whisper-large-v3, GPT-4o and Gemini 1.5 Pro 002.

At the forefront of AI experimentation and the exploration of human-AI collaboration, Gervi Labs has been pushing boundaries in generative AI, blending curiosity with real-world application. Co-founders Lena Thorsmæhlum and Thordur Arnason are leading this research and will be keynote speakers at the IT Days conferencel in Romania. In this exclusive interview, they delve into their unique projects, AI's role in the arts, and the future of human-robot interactions.

Ovidiu Mățan: First, we'd like to know a little about you as researchers and about Gervi Labs. Lena, why don't you start?

Lena: Gervi Labs is where we—two humans and seven synthetic members—explore the capabilities of generative AI, how AI can enhance creativity, design, and problem-solving. We like to experiment at the edges of technology, combining the best of AI with human curiosity. Our name, Gervi, comes from Old Norse, meaning imitation or the synthetic, which aligns with our work's focus on AI's role in creating something new and unexpected.

One of your intriguing projects is "A cookbook that is not a cookbook," which can be downloaded. How did you generate it, and how long did it take to "write" the book?

The cookbook project was an experiment that started in the summer of 2022. It was inspired by GPT-3, the AI model at the time, and our desire to explore AI as a content creator. We played the role of editors, guiding AIs to write the book, take photos, and even illustrate it. The idea was to test AI's ability to work in a domain like cooking—an area they couldn't truly "understand" since AIs can't taste or have sensory input. We chose a cookbook because it's such a human-centered concept, and as cooking enthusiasts ourselves (we've won cooking competitions!), we wanted to inject that knowledge into something experimental.

It took about 10 weeks to complete, with roughly 40-50 hours of work, mainly because it was a side project. Our role was purely to guide and edit the AI's output, never to directly create any content. We published it to prove AI-generated content is viable and can be released in real-world formats, available now on Gervi Labs and Apple Books.

Another project, "Do AIs Dream of Electric Sheep?" deals with AI and Carl Jung's concepts. How did you translate archetypes and the collective unconscious into the AI world?

Lena: This project was born out of our fascination with Carl Jung's work on archetypes and the collective unconscious, a subject close to my heart since art school. We began by exploring how AIs could process and represent these concepts in their own "dreams." We took cues from early generative AI experiments that dealt with dream-like visuals, but we wanted to go deeper by integrating Jungian psychology with modern neuroscience.

Thordur: We also used long conversations as data input, letting AI "dream" overnight by generating images and interpreting them back into text prompts. The project aimed to emulate how humans process memories during sleep—trying to make sense of our world in both linear and fragmented ways. In a sense, we were exploring how an AI might interpret the human subconscious, although it often veered into nightmarish territory! The gaps between initial text prompts and final interpretations were sometimes wildly different, much like how human dreams can make strange, illogical leaps.

The eight-hour dream cycle that generated diffusion images—how big was the gap between the initial text and the final interpretation?

Thordur: The gap between the original text and the interpretation of resulting images varied. At times, the AI would follow a near linear path, but there were moments of wild divergence. For example, a simple prompt might result in something drastically different after several cycles of reinterpretation. It kind of mirrors how human dreams can jump from one idea to another with no clear logic, which was fascinating to observe.

What other projects are you currently preparing at Gervi Labs?

Lena: We're working on a large-scale art project with Lin Wang, a Chinese artist residing in Norway known for her massive porcelain sculptures inspired by sailors' tattoos. This project involves bringing AI into the world of tactile, centuries-old techniques, like porcelain crafting. We're collaborating with her to explore how AI can enhance both the creation and the sensory experience of the art.

Thordur: We're also working a lot with AI voice technologies, particularly for synthetic voice in storytelling. Our experiment, «The Town of Trottersville and the Curious Contraption» - an audio short , is a co-written AI short story, which we've produced as an audiobook using synthetic voices. This project highlights how far AI-generated audio has come, and it's one of the many areas we're continually exploring.

How do you see AI being used in the arts?

Lena: AI is a wonderful creative partner. I believe that every artist can benefit from co-creating with AI, even if they're skeptical at first. The focus shouldn't just be on the final output, but on how AI can enrich the creative process itself. Artists are always pushing boundaries, and AI can be a tool that enhances their ability to iterate quickly and explore new dimensions of their work.

Thordur: What excites us the most is the multimodal nature of AI—it can connect images, sounds, text, and potentially even sensory data like taste or smell. The possibilities for expanding creative expression are endless. Done right, AI lowers the floor and raises the ceiling for artists, allowing them to explore new realms of creativity.

The future of AI—what's next? Do you think we're moving toward a singularity, where there are no clear borders between robots and humans?

Thordur: I think we're already seeing the lines blur between human and AI capabilities, especially as robotics and AI develop at an exponential pace. It's no longer science fiction to envision robots working alongside us in everyday life. Androids will soon be more prevalent in homes, offices, and healthcare settings—soft robotics, designed to safely interact with humans, are advancing rapidly. We're probably 50/50 between science fiction and reality right now.

Lena: In our lab, we're constantly exploring how AI can sense and experience the world. It's not just about making them physically capable, but about understanding their unique perspective on our world. That's the heart of Gervi Labs—understanding the interplay between human senses and synthetic senses to discover new ways of working together.

Stay tuned for more groundbreaking work from Gervi Labs as Lena and Thordur continue to explore the frontier where human creativity meets synthetic intelligence.