As AI has become more accessible, we have witnessed examples illustrating its diverse applications. Prominent among these are generative AIs, which excel in their ability to “create” images through prompts, many distinguished by their composition and vividness. These AI systems are neural networks with billions of parameters, trained to create images from natural language, using a dataset of text–image pairs. Thus, although the initial question posed by Turing in the 1950s, “Can machines think?” still recurs today, the generation of images and text is grounded in existing information, limiting their capabilities.
What has surprised many is the increasingly apparent closeness to overcoming the Turing test and the growing similarity, in terms of visualizations, to what an architect with skills in this field can achieve. In this context, while the debate persists in the architectural community about whether AI can process architectural concepts, this article explores how it interprets materials to develop these visual representations. With that in mind, a single prompt was developed for this experiment (with materiality as its variable) to delve into the obtained results.
Prompt: A contemporary building immersed in a built environment. This building is made of (material). An isometric view with realistic textures.
The generated visualizations may imply that the AI interprets the building within its context and understands various applications of the materials. Thus, some of them depict a contextualization of materials, emphasizing their use as part of a construction system. This aligns with common applications in specific building types and introduces a stylistic element to the process.
The prompt led to the creation of visualizations, organized by material types commonly used in architecture. These materials, including brick, steel, adobe, wood, concrete, and stone, were chosen for their distinct characteristics, facilitating comparisons.
Brick
With this material, a common thread was the robustness of the buildings, linked not only to the properties of brick but also to its technical characteristics that do not favor the addition of large windows on the facade. It is noteworthy that AI recognizes the variations of the brick in terms of shades and textures, making its aesthetics visible. In this way, the visualizations could suggest projects of various typologies, ranging from collective housing and public buildings, such as Exeter Library, to industrial facilities like the Fagus Factory.
Steel
The results of this material are exemplary in showcasing an "understanding" of the role of steel. Instead of employing it as the dominant texture on the facade, the visualizations suggest its application as a construction system, often in conjunction with glass facades or what appears to be ventilated facade systems. In this manner, certain buildings exhibit design elements reminiscent of projects like the Seagram Building and public spaces like Centre Georges Pompidou.
Adobe
With Adobe, although the prompt refers to a contemporary building, features that evoke regionality can be observed, generating visualizations with a local character. These features include small rectangular windows, a pitched roof similar to Earth House, and the involvement of textures that might suggest wood. The Plúmula Workshop House project exhibits similarities with the obtained results in the tones of the building and its geometry, as well as in the presence of a patio terrace at the top.
Wood
In the visualizations, the consistency is evident in that none of the materials suggesting wood have the same shades. This color spectrum varies from light tones, possibly reflecting the appearance of plywood, to darker textures such as walnut. Some images exhibit structures of two or more levels, and one specific image presents an elevated structure similar to those found in humid climates, as in projects such as the Boca de Agua Hotel, as well as volumes similar to Open Park Villa.
Concrete
The concrete results reveal visible combinations with glass facade systems, reminiscent of the long windows and open facade of Gropius House. Of all the images obtained, these volumes are the ones with the largest number of blind faces on the facade. In some cases, this gesture is an essential part of the building's identity, as seen in the case of Folm Arts.
Stone
Stone proved to be one of the most challenging materials to represent, given the more perceptible differences between the visualizations. Several reasons could account for this difficulty, ranging from the challenge of AI to represent materials sourced from nature —similar to the challenges faced in representing hands— to the diversity of types of stones. The shape of the volumes suggests residential projects similar to Viglostasi Residence, as well as integrations of pre-existing elements, exemplified by the Hacienda El Barreno Visitors Pavilion.
In comparing artificial intelligence with human development, we are still in the "childhood" of this technology. It will be fascinating to observe how it evolves in the future and how the understanding of AI transforms to interpret abstract and artistic concepts, especially those linked to architecture.
So far, this experiment suggests that the AI recognizes that the character of some buildings is linked to the use of specific materials, establishing a direct relationship with typologies. Despite the prompt being designed using the word "building", the AI presents visualizations that suggest a broad range of structures, including single-family housing, industrial complexes, and multi-story buildings.
Although it is undeniable that AI can distinguish between the textures of materials, it seems there is still a way to go before it can generate complex proposals that arise from a deep understanding of the built environment, materials, their characteristics, and, most importantly, their essence. But perhaps, in a few years, the title of this article will evolve from a question to a statement.
