by Olimpia Bellan 

Joshua Vermillion, AI whisperer, architectural provocateur, and digital craft visionary. A tenured architecture professor and award-winning designer turned AI explorer, he’s been quietly building entire worlds through generative models, spaces that look like they belong to tomorrow, but echo the logic of real architecture. His work blurs the line between machine imagination and human intention, turning spatial dreams into visuals that feel vividly tangible. We asked him what happens when architecture meets AI.

Your AI images feel like buildings from a dream, fluid, impossible, yet strangely believable. When you create these spaces, do you think of yourself as designing architecture, or as inventing worlds?
I enjoy world-building and architecture that are centred around narrative and storytelling. Often times, I feed in my own design work (photographs, drawings, models, etc) to the gen-AI model and the results feel derivative of the old work but with a different, almost alien, quality to them. I’m most interested in exploring and interrogating this “otherness” resulting from the machines’ very limited understanding of our world.

You’ve generated thousands of imagined spaces with AI. Has working at that scale changed how you understand the way people instinctively read space, form, or atmosphere?
Using AI has definitely let me explore the use of scale/size, material qualities (textures, translucencies, etc), and atmospherics (lighting and other effects) as a given. One can be very prolific in exploring a set of ideas very quickly, and posting results on social media has also given me some insights (almost in real time) on how different spatial and material decisions are perceived. AI’s speed almost perfectly complements the iterative brainstorming at the beginning of the design process.

Your work blends computational systems with hands-on craft. When AI generates a scene, where do you draw the line between the machine’s creativity and your own?
First, can a machine really be creative? Let’s say, for the sake of argument, that machines can appear to do creative things even if the model doesn’t really know why we would consider it creative. In reality the model is combining my imagination with its generative latent space and sometimes it gives back results that I would have never thought of. However, while you can delegate some of the execution (image-making) to the machine, you can’t delegate judgement, critical thinking, and other inherently human qualities that are necessary when designing.

There’s a sculptural quality to your AI visuals, structures that feel like they could be built, but also feel like they’re evolving. Do these images influence your real-world architecture work, or are they a separate creative universe?
Tangible work is still a very (human) labour-intensive process. Whereas, the images from gen-AI are very facile, superficial or naïve. I don’t ever ask gen-AI for truth, but rather, it’s largest affordance seems to be iteratively exploring possibilities with material or spatial effects, atmospherics, with astonishing speed. This is the part that can be very helpful at the beginning of the design process when divergent thinking is most valuable.

Your work suggests AI is more than a tool, it’s a catalyst for new ways of seeing. As your art has evolved alongside generative systems, what has AI taught you about your own creative process?
I think the most fascinating thing to me is that we now have ways to let machines learn (albeit in a very narrow sense currently). And when communicating to a machine as a design “partner,” one has to internalise and then orchestrate the process, not just the ideas driving the process. I think in the future architects and designers, augmented with AI, will have to shift more to designing processes and workflows, not just designing buildings.

Read Joshua Vermillion interview on BMI:MAG limited edition, find your copy at @bluemarlinibiza and throughout Ibiza’s hotspots