Can AI mark the next Architectural Revolution?
By Karla Saldaña Ochoa
The evolution of computation has led to a decentralized shift from singular functional machines to a world of applications. We have moved from one central computer serving many people (1943) to one computer per person (1974) and now to one person using multiple devices (2023). These devices constantly gather information about our daily activities. They are omnipresent, creating vast amounts of data through actions such as walking, purchasing, traveling, and even entertainment consumption. This data, referred to as big data, comprises various formats such as images, videos, and texts; it is multimodal, and we have a lot of it. However, Artificial Intelligence (AI) algorithms can convert these digital data into a numerical form for analysis. Using these numbers to make predictions, we can find patterns and make decisions.
The use of the term “Artificial Intelligence” can be traced back to Alan Turing, who introduced the concept almost 70 years ago. Turing proposed that machines can exhibit intelligent behavior by tricking people into thinking they are human; he tested his hypothesis on a question-and-answer game that he called the Turing test. In the test, an interrogator tries to determine who is a human and who is a machine based on their answers. If a machine is mistaken for a human, it has passed the Turing Test. Adaptation to this question-and-answer test has been adapted to many tasks, such as chatbots, robotics, and image recognition. Industry and academia have been competing to pass those “Turing tests” with their AI systems for years. This competition has led to a diverse toolkit of AI technologies that researchers and designers can choose from to achieve their goals. As this article is about artificial intelligence algorithms and architecture, let’s start by first defining what generative algorithms are (the most common algorithms used in computational design) and where they sit within the broader umbrella that encompasses AI.
To summarize, generative algorithms generate new data by modeling the underlying probability distributions in existing data. Some algorithms are AI but not Machine Learning (ML), and some are ML but not deep learning (DL). All these terms have been used interchangeably to refer to the same concept (an intelligent system); however, they are different. AI is the umbrella term that houses ML and DL. For example, a rule-based generator is an AI algorithm but not an ML algorithm, e.g., a parametric algorithm. An example of a DL generative algorithm is a diffusion model (a part of the well know algorithms, mid-journey, and DALL-e), as it learns from data and has an architecture with many layers (therefore deep). A diffusion model aims to learn the latent structure of a dataset and model how data points diffuse through the latent space. Today we will discuss how they are and can be in architectural design.
Recently, we have seen a rise in images and text produced with generative AI algorithms using the DALL- e, Midjourney, and ChatGPT. They are so fun to play with, because, with a simple input text, a newly generated output is generated. Because of those outputs, we also have seen a rise in skepticism of whether such agility in production threatens a designer’s job.
I recently attended a gallery exhibit and panel discussion about AI in architecture and witnessed how easily architects embrace the idea of image production. I saw nearly 300 images and 100 AI-generated text descriptions for those images. When architects presented their AI-generated work, it felt that the architect’s role was to curate images and select a more formally appealing sample. When we entered the discussion of the panel, the first question was: Who is the author of those images and text descriptions? Is the architect that prom the AI generative algorithm? I would argue that they are not… precisely. Those are not their projects per se but are iterations of possible projects that they have curated to inspire future projects — the authorship is shared with the researchers that create the algorithm, the people that produce the training data, and the designers that generate the images. And here is a warning, we should be careful in assigning all the design responsibilities to a machine, thinking that it is a recommendation engine, similar to the ones that recommend the next movie to watch.
Let’s dive into another medium, text. Text allows designers to describe the design space and present ideas they (designers) are exploring or have yet to explore. For this let’s use ChatGPT. For the sake of experimentation, I prompted it with the following text, “write an architectural project similar to one of a Le-Corbusier project“ after a few seconds, the following text appeared:
“This project is an architectural marvel that transcends time and trends, staying true to the principles of modernism espoused by Le-Corbusier. The building boasts an iconic silhouette characterized by its sleek lines, geometric shapes, and innovative use of raw materials such as concrete, steel, and glass. The open floor plan and large glass walls offer an unparalleled connection with the outdoors, blurring the line between interior and exterior spaces and providing ample natural light. Sustainability and environmental responsibility are key elements of this project, with a strong emphasis placed on the use of greenery and vegetation to create a harmonious balance between the built and natural environments. The minimalist color palette and industrial feel give the building a raw, modern edge, while the strategic placement of vegetation provides a touch of nature to the urban landscape. In short, this project is a contemporary interpretation of Le-Corbusier’s timeless design principles, offering an unrivaled living experience in a harmonious balance between form and function. Get ready to experience the ultimate fusion of style, sustainability, and innovation.”
Where is the creativity in this process? Writing an architectural brief is not about listing the program but synthesizing the design thinking translated into research about the site, comfort, and users. It also connects to the future user at the level of their needs. Moreover, I wouldn’t say that the above text aligned with Le-Corbusier’s work. He didn’t focus on sustainability and environmental responsibility, inside-outside relationships, etc.
With this article, I want to turn the attention to how AI can help architects and designers make better decisions, generate new ideas, and explore design spaces that were impossible before. Architecture is not only about curating images focusing on form and program description. But an architectural project synthesizes many layers: form, functionality, need, and comfort. For about 100 years, the way we design, and construct buildings have mostly stayed the same, which is not the case for planes and aerospace ships. To move forward, we should acknowledge the tradition of producing architecture but embrace the new paradigm that AI and data-driven design methodologies bring to architectural design. We can now modify how architects design because they can access millions of design solutions sorted by AI algorithms based on the architect’s preference or biased generators to follow a particular design prone to satisfy the designer’s style. AI could be used as a tool to support, not replace, the creativity and expertise of architects and designers.
To make my point, I will show a project where AI was used to organize thousands of images from social media from users describing their needs for a particular project. Not being able to be physically in a place, the only source of information in such a place was social media data. The total amount of images and posts collected where more than 12,000. How to make sense of this plenty of data? By using AI algorithms, we created an image that resembles a common ground where a designer can understand the overall feeling of the users about the place. This was only possible by using AI algorithms. A designer can depict a few images from the common ground to create atmospheric images. These images attempt to capture the designer’s intention through concepts she wanted to engage in her design. Through this exploration, an architect can understand the site differently, including a collective viewpoint characterized by the users and their activities. This approach goes beyond the typical site analysis that relies on limited site visits and mappings related to the site’s physical infrastructure and targeted interviews.
Let’s look at another example. Here are some atmospheric images accompanied by text; this time, for image generation, the designer combined the previous common ground and DALL-e, prompting the generator with concepts he wanted to embrace in his project. He used ALICE, a web-based AI search engine, to write the design brief for his projects. Alice has direct access to libraries of architectural treatises. The aim of this project was to use an AI-powered tool to write a design brief that challenges the standard individualistic approach. Now, the discussion is enriched by various authors addressing the same topic to articulate a comprehensive description of the design intentions. It is like collaborative writing; the only difference is that your co-authors are renowned characters that existed 100 or 1000 years ago.
Here is my take-home message using AI to find a design solution does not stop the design process; by no means it just begins; it is like having an active and responsive precedent suggestion tool. I have shown you examples that are purely 2D; however, now I will introduce you to some examples where other layers of information are included thanks to working with AI algorithms. This idea is developed in detail in this article, co-authored with Patrick Ole Ohlbrock, Pierluigi D’Acunto, and Vahid Moosavi.
Today AI is challenging the role of human designers by creating end-to-end generators capable of infinite solutions. However, we need to measure the effectiveness of the solution (even infinite solutions) against the relevance of the question it addresses, understanding that AI’s potential can only be achieved when it is actively guided. Recently, researchers focused on using AI for end-to-end generative processes that learn from massively crowdsourced online datasets to create realistic images (2D) that can be automatically generated from text inputs.
However, design tasks hardly fit into a 2D framework and, instead, fall into n dimensions representing qualitative aspects such as user-specific preferences, perception of space, etc., which need to be adequately formulated as well-defined design objectives. To come back to one of our initial questions of curing versus enriching their work, Architects should use AI to add many complex layers to their design going beyond the traditional two-dimensional representation to n-dimensional representation (beyond 2D by adding additional layers of data that encapsulate site requirements, comfort, measurements, etc.). Imagine an assistant trained to learn your design preference and acts as a sounding board where your ideas mature and develop. We are getting there, and today, generative algorithms are helpful tools. Still, they need to evolve to consider human feedback that includes those extra layers of information that ultimately create an architectural project.
Recently we proposed an AI generative framework that aims to strike a balance between the efficiency of AI and the critical and creative thinking of human designers. By incorporating descriptive input to guide the AI-generated 3D designs toward meeting the desired design objectives (structural performance) while also capturing the qualitative aspects essential in the architectural design process. This project aims to address the current limitations of AI-generated designs and provide designers with a powerful tool to tackle the complex design challenges facing our natural and built environment. If you want to know more about the project, here is the link to our initial publication (https://www.ai-share-lab.com/text-2-form-3d). We are currently working (with Zifeng Guo and Pierluigi D’Acunto) on the new version, which will be published by the end of 2023.
When designing, architects try to balance novelty and familiarity. On the one hand, we desire novelty, as seen in fields like fashion, but we also like what we are familiar with. With AI’s help, we can find a balance between novelty and familiarity. The opportunity that AI brings to the profession is just starting. AI generative algorithms are valuable instruments, but as with any other instrument, you must be proficient enough to tune them. The outcome is as much as the artist as the algorithm. Because not everybody can play the piano like Martha Argerich or write novels like Mark Twain, they both used instruments in tandem with human mastery
AI won’t take our jobs; it will augment our search, design, and production capabilities. The AI I envision can adapt to the personal preferences of designers and act as a suggestion engine for the semantic interpretation of architectural projects. For me, AI is not only for solving well-defined problems but tackling weakly defined tasks inherited from the creative nature of the human designer. AI algorithms can generate images, but they still lack the human touch, creativity, and cultural sensitivity that a trained architect possesses. The role of architects will not become obsolete as AI cannot replace the critical thinking, empathy, and intuition that are essential to the design process. AI can be a tool or instrument for architects to aid in the design process and generate new ideas, but more is needed to replace the human element in architecture. Our training in architecture gives us a deep understanding of design, history, culture, and technology, and these skills will always be in demand. However, we must be vigilant about how AI is used in the field and never let it replace her creativity and intuition.
About the Author
Karla Saldana Ochoa is a Contributor and Advisor at Urban AI. She is a Tenured Track Assistant Professor in the School of Architecture at the University of Florida, and a faculty affiliate at the Center of Latin American Studies and FIBER, the Florida Institute of Built Environment Resilience at the University of Florida. Her teaching and research investigate the interplay of Artificial and Human Intelligence to empower creativity and social good. Karla leads the SHARE Lab, a research group focused on developing human-centered AI projects on design practices. Karla is an Ecuadorian architect with a Master of Advanced Studies in Landscape Architecture and a Ph.D. in Technology in Architecture from ETH Zurich. Her Ph.D. investigated the integration of Artificial and Human Intelligence to have a precise and agile response to natural disasters.
