As artificial intelligence becomes increasingly embedded in architectural production, much of the conversation surrounding it has focused on speed. Generative models can produce images in seconds, rapidly visualize ideas, and multiply design options at a scale that would have been difficult to imagine only a few years ago. For architect and designer Alper Turan, however, this acceleration is not necessarily the most interesting consequence of the technology. His research project The Artificial City approaches artificial intelligence from almost the opposite direction: as the beginning of a deliberately slow process of architectural interpretation, reconstruction, and drawing.
Rather than treating these inconsistencies as mistakes to be eliminated through better prompting, Turan takes them as the starting material for design.
AI as a Generator of Architectural Ambiguity
Rather than asking AI to resolve a building or visualize an existing design, Turan uses generative models to produce unstable interpretations of familiar architectural types. Churches, mosques, synagogues, castles, train stations, schools, pump stations, McMansions, and neoclassical buildings become prompts through which the model attempts to reconstruct architecture from the vast visual archive embedded within its training data.
This process generates hundreds of images, producing objects that are simultaneously recognizable and unfamiliar. Architectural characteristics persist, disappear, combine, and mutate. The resulting images often appear convincing at first glance but become increasingly strange when examined as buildings.
This instability is precisely what makes them useful. Generative AI can reproduce the appearance of architectural categories while remaining largely indifferent to the spatial, structural, and organizational principles through which architecture actually operates. A train station might possess the monumentality and iconography of a station without establishing a plausible relationship between platforms, circulation, structure, and interior space. Rather than treating these inconsistencies as mistakes to be eliminated through better prompting, Turan takes them as the starting material for design.
From Image Generation to Architectural Interpretation
From this point onward, the process moves largely outside the generative model. Turan selects particular images and fragments from the larger archive and begins reconstructing them architecturally through plans and three-dimensional models. Forms that initially existed only as pixels must acquire thickness, circulation, rooms, entrances, structure, and spatial relationships. The apparent completeness of the AI-generated image begins to dissolve as soon as it is subjected to the requirements of architectural representation.
This translation is intentionally interpretive rather than systematic. No automated procedure turns an image into a building. Some generated forms are reconstructed relatively closely, while others are altered, exaggerated, combined, or abandoned altogether. Ambiguous areas become opportunities for invention, while elements originating in separate images can migrate into a single architectural object.
The process consequently shifts authorship away from the initial act of generation and toward a prolonged sequence of selection and transformation. AI operates less as an autonomous designer than as a source of unfamiliar propositions, an external imagination capable of introducing forms and associations that might not otherwise appear within a conventional design process. Its lack of architectural understanding, rather than being an obstacle, becomes productive.
Drawing After AI
Drawing ultimately becomes the mechanism through which these propositions are claimed and transformed. After developing the selected buildings through plans and models, Turan reconstructs them again as large architectural drawings. Many are represented through unconventional worm’s-eye perspectives that expose their undersides, transforming plans, circulation systems, structural fragments, and architectural details into dense graphic compositions.
The decision to redraw the architecture is particularly significant at a moment when AI can generate increasingly polished architectural imagery almost instantaneously. Rather than accepting the generated image as the final artifact, Turan deliberately introduces additional stages of labor between generation and representation. The process moves from prompt to image, image to selection, selection to model, and model back to drawing.
What begins as an automated process consequently becomes increasingly manual. The final image is not a refinement of the original AI output but the result of reconstructing the architecture behind it.
Using AI Slowly
In this sense, The Artificial City proposes an unusually slow use of a technology defined by acceleration. The abundance of images generated by AI makes selection more important rather than less. Their ambiguity requires interpretation, while their spatial inconsistencies create opportunities for architectural decisions.
The machine produces quickly, but the architect works slowly.
This reversal distinguishes the project from workflows in which generative AI primarily functions as a visualization or optimization tool. Turan instead uses the technology to introduce uncertainty into the design process. The value of the generated image lies precisely in the fact that it cannot be entirely controlled. It offers an architectural proposition without fully understanding what it has proposed, leaving the architect to determine what can be extracted from it.
The apparent efficiency of AI is therefore interrupted by a deliberately inefficient process of looking, choosing, modeling, and drawing. Rather than shortening the distance between an idea and its representation, The Artificial City extends it.
Constructing the Artificial City
As the collection expanded, the individual experiments began to form a speculative urban environment. Mosques, mansions, schools, castles, train stations, religious buildings, infrastructure, and other familiar typologies accumulate without the organizing logic of a conventional master plan. Instead, the city derives its coherence from a shared method of production. Each building begins with an artificial image and passes through a similar sequence of interpretation, modeling, and drawing before entering the larger collection.
The resulting city is artificial in more than one sense. It is produced from images generated by artificial intelligence, but it is also assembled from distorted memories of architectural types. The model draws from a vast collective archive of existing architecture, producing approximations of what a castle, mosque, mansion, or station is supposed to look like. Turan then reads those approximations architecturally, turning the model’s visual assumptions and misunderstandings into spatial propositions.
Familiar typologies consequently return as strange descendants of themselves. Together, they construct a city whose architecture originates neither entirely from the machine nor entirely from the architect, but from the repeated translation between the two.
Who Authors an AI-Assisted Drawing?
Underlying The Artificial City is a broader question of authorship. Generative AI complicates the traditional relationship between architect and image because the person initiating an image no longer determines every element that appears within it. Turan’s project does not try to resolve that ambiguity by minimizing AI’s role. Instead, it asks where authorship might lie in a process that begins with machine-generated material.
Rather than shortening the distance between an idea and its representation, The Artificial City extends it.
Physical Manifestation
The Artificial City ultimately moved beyond the screen and the drawing into a physical exhibition at Tulane University School of Architecture and Built Environment. Presented at Richardson Memorial Hall in Fall 2025, the exhibition brought together fourteen large-format drawings and ten physical models, translating the project’s digital experiments into a spatial installation. The models gave physical form to architectures that had begun as ambiguous AI-generated images, while the drawings surrounded them as both architectural documents and autonomous visual artifacts.
About Alper Turan
Alper Turan is an architect, designer, and educator whose work operates across architecture, graphic design, and visual culture. He earned his Master of Architecture II from the Yale School of Architecture and worked as a designer at Pentagram in New York, where his work spans environmental graphics, identity, and the intersection of architecture and communication. From 2024 to 2026, he served as a Visiting Assistant Professor of Architecture at Tulane University School of Architecture and Built Environment, teaching architectural design and representation while developing independent research, including The Artificial City. His work has been published and exhibited internationally. Alper continues his independent research and professional practice through Post Form, working across architecture, design, and visual culture.
