Frontiers of Artificial Intelligence in Architecture pt.1
Intelligent Artifices? A Critical History of (and New Directions for) Generative Design
Thursday
18th April 2024 | 12:00 – 14:30 CET
Aula 201 Lingotto | Streaming on YouTube
INTRODUCTION
Despite the aura of novelty that surrounds artificial intelligence (AI) methods in architecture and design, dreams of autonomous design machines have long populated the imagination of technically inclined architects, designers, and artists — and that of artistically-inclined mathematicians, scientists, and engineers. What do these ambitions tell us about the interplay of technological and cultural understandings of design, and how have they shifted with the recent explosion of data-intensive “generative” computation?
In this talk Prof. Cardoso Llach will explore the long history of generative design methods, and consider some critical issues concerning the recent adoption of machine learning techniques in architectural contexts.
Finally, Prof. Cardoso Llach will sketch out some new directions for employing these in design, architecture, and construction drawing from recent research at Carnegie Mellon University’s Computational Design Laboratory.
BIO
Daniel Cardoso Llach is Associate Professor of Architecture at Carnegie Mellon University, where he chairs the graduate program in Computational Design and directs CodeLab, a research and learning laboratory rethinking the role of computation in the design and production of the built environment. He is the author of publications, exhibitions, and technologies critically exploring the nexus of design and computation, including the book Builders of the Vision: Software and the Imagination of Design, a history of computer-aided design (CAD) that identifies and documents technological theories of design emerging from postwar government-funded research projects at MIT, and reflects critically on their architectural repercussions. His new book, Designing the Computational Image, Imagining Computational Design, co-edited with Theodora Vardouli, draws from historical and contemporary materials to trace the emergence of computational design ideas and practices across a broader landscape of institutions in the US, the UK, and Canada.
Contacts
The lecture was not only inspiring but also remarkably comprehensive in its exploration of AI capacities. The lecture provided a comprehensive overview of the integration of AI in architectural design processes, shedding light on both its capabilities and limitations. The speaker effectively navigated through various stages, from AI brainstorming to custom model training, showcasing how AI can augment and refine architectural workflows. The emphasis on utilizing AI as a tool for creativity and iteration rather than a replacement for human ingenuity was particularly insightful. The lecture underscored the importance of critical thinking in evaluating the coherence and constructability of AI-generated designs, urging architects to balance technological advancements with practical considerations. Additionally, the discussion on the evolution of AI in architecture, from initial hype to realistic productivity, provided valuable context for understanding its current role and future potential. Overall, the lecture sparked curiosity and encouraged further exploration of AI’s possibilities in architectural practice, while also emphasizing the need for responsible and mindful application.
In his opening remarks, speaker Daniel Cordoso Llach provided a captivating journey through the annals of generative design, tracing its roots back nearly half a century. His insights shattered the misconception that artificial intelligence in design is a recent phenomenon, shedding light on its early beginnings. As he delved into the presentation, Llach unveiled a myriad of applications for generative design, showcasing its versatility and potential. Transitioning seamlessly, he delved into the nuanced complexities surrounding this technology, touching upon pertinent issues such as citation practices, carbon footprint concerns, and the often-overlooked labor involved in its development. Exploring the evolving landscape of AI in architectural design, Llach introduced three pivotal conceptual moves: interface, hybridity, and materiality, offering a glimpse into the future possibilities for designers. His vision of AI’s role in design was both thought-provoking and inspiring, sparking contemplation on the symbiotic relationship between human creativity and technological innovation. As he concluded, it became evident that the intersection of AI and design holds boundless potential for shaping the world of tomorrow.
In his presentation, Daniel Cardoso Llach explored how Artificial Intelligence (AI) is impacting architecture. He noted that while AI isn’t new, recent advancements in computing have made it more powerful. Llach pointed out that much of today’s AI focuses on 2D pixel-based technology, but he advocated for a shift towards 3D-based approaches to enhance construction processes. He also raised concerns about the significant energy consumption associated with AI and stressed the importance of developing sustainable alternatives. Llach emphasized the need for AI applications to be grounded in practical, real-world contexts to maximize their effectiveness. Overall, his presentation underscored both the potential benefits and challenges of integrating AI into architectural practice, highlighting the importance of thoughtful and sustainable implementation. Llach discussion on the integration of AI in architecture can be related to the conference’s general topic “the limits of architecture”, as it challenges traditional boundaries while also highlighting the importance of ethical and sustainable practices.
As a student attending the AI lecture, I was captivated by the profound impact artificial intelligence can have on architectural design processes. From the initial stages of project inception to the final documentation phase, the possibilities presented by AI are truly transformative.
The concept of AI brainstorming opened my eyes to a new realm of creativity, where ideas can be generated on the fly, pushing the boundaries of traditional design thinking. The collaboration between MVRDV and Midjourney showcased how AI can facilitate collective dreaming, leading to the creation of visionary architectural concepts.
The notion of AI as a “Super Pinterest” resonated with me, highlighting its potential as a reference generator to inspire and inform design decisions. Additionally, the idea of using AI to combine disparate concepts into cohesive solutions was both fascinating and practical, offering a fresh approach to design development.
Exploring AI-generated collages as a tool for schematic design was particularly intriguing. The ability to test ideas without extensive modeling provides a valuable opportunity for rapid iteration and exploration, ultimately accelerating the design process.
Looking ahead, the concept of customization through AI holds immense promise. The idea of developing custom AI workflows tailored to specific design needs, as well as integrating AI into live design sessions for real-time feedback, opens up exciting possibilities for innovation and iteration.
Overall, this lecture has sparked my imagination and deepened my appreciation for the intersection of AI and architecture. I am eager to explore these concepts further and harness the power of AI to push the boundaries of design creativity in my future projects.
In Fredy Fortich’s lecture on AI in architecture, he outlined a journey of exploration and integration of AI tools into architectural workflows. He emphasized the use of stable and reliable software platforms such as Rhino and Photoshop as the base for incorporating AI techniques. For drawing desired elements, he suggested utilizing base PSD or base sketches with high denoising strength and low denoising for optimal results.
Fortich discussed various applications of AI in different architectural tasks, starting with massing studies where a stable base render is employed. For interior design, he recommended using clean renders along with Enscape channels and providing key words for generating desired outcomes.
When it comes to materials, Fortich suggested using low-quality wireframes along with text prompts to guide AI in material selection. He stressed the importance of understanding the character of an image and incorporating textual descriptions to enhance the narrative.
In the context of upscaling images, Fortich discussed custom upscaling techniques using software like SD and IPScale to achieve high-quality results. He also introduced the concept of live AI, where custom LORAs (Local Operation Reinforcement Agents) are employed for interior design tasks, similar to the approaches used by renowned firms like MVRDV. The combination of stable diffusion and custom LORAs leads to the generation of new architectural concepts.
Fortich highlighted the significance of reinforced learning through human feedback (RLHF), which involves pairing text descriptions with images to refine AI-generated designs. He advocated for the use of separate LORAs for each project, allowing for tailored AI assistance based on specific design requirements and objectives. Overall, Fortich’s lecture provided valuable insights into the integration of AI technologies into architectural practice, offering practical strategies and considerations for architects looking to leverage AI tools effectively.
Prof. Daniel Cardoso Llac talks about general historical evolution of AI, he specifies that AI is a term that include many tools, it always sounds like something ominous, it feels like something bigger then ourself.
AI is already used to help in some architectural project, but professor Cardoso Llac emphasize that the idea of automated machine and automatic design, it’s been around for a time.
Nowadays he suggests three ways to think at AI in a new direction, three conceptual move.
The first is about data material, which is what we use to teach Ai what product we want back, the second is about hybridity, from an interdisciplinary point of view; and last but not least he question the interface between human and machine, because, often, as architect we are often user of technology, but we should think of ourself as makers.
In the end, he underline some different critical problems, such as ethical and environmental, but also that we should take into consideration that what we see from a product of AI is always pixels based.
The lecture had thrown light on the aspect that for such a long time, the world of computing, parametricism had been existing in multiple forms. The Artificial intelligence that seemed to be having a boom these days is nothing but an advance version of these technologies already persisting in the design industry. Artificial Intelligence has created a realist utopia. One very alarming knowledge gained through the presentation as about the relation of the growth of artificial intelligence with the carbon footprint. Each AI generated image is costing the environment way more than the standard usage of any software or technology does. It is therefore also vital to question the growth of this industry from environment point of view. There were some good suggestion of the literature by the prof related to climate and technology and other dimensions involved like Climate Vs capitalism. Overall Architecture and technology is an absolutely growing direction that we all must pay sincere heed to.
It was an informative lecture starting with an overview of prehistory generative design with AI followed by the contemporary techniques associated with it. Professor Cardoso explained how despite seeming new, the idea of automated machines has been around for a while. The idea sort of started with the works of Brooks and his collaborators in the 1950s, who developed a method to generate musical sequences by numerically encoding a bunch of musical pieces. Architectural researchers soon got interested in the world of computer art, specifically what is called “shape grammar” which is a generative design method used in architecture, urban design, and other creative fields for creating complex designs by defining a set of rules or grammar that govern how shapes or forms can be combined and transformed.
Then we have the emergence of more conversational ways of communicating with computers; programs in which you can speak to the system by typing. For example, you tell the computer that you need this amount of entrances and the system generates an option.
The origin of today’s machine learning techniques is rooted in some inductive reasoning ideas, statistical analysis and probability, and a computational interpretation of actual biological neural networks. So the technology is not new, it is from the 1950s, but what has happened during the last 15 years is access to more computational power and large data sets.
Professor Cardoso also pointed out some critical elements of using AI, first of which is ethical issues. As there is so much labor behind any pic on the net, the copyright and the label should be regarded. This issue can be resolved with the same approach as MVRDB. The second issue is the environmental impact and the energy consumption to process pics on the net, which again can be reached by limiting the domain of the process to just pics on the company’s property. And the third challenge would be moving away from utopianism to the ideas more rounded to reality.
Professor Cardoso’s lecture on the critical history and future of Generative design explored its evolution, pivotal concepts, and potential trajectories. He began with Palladian grammar, highlighting Generative design as a platform for diverse viewpoints. Referencing Ivan Sutherland’s “Sketchpad,” he advocated for utilizing computers’ capacities in design, shifting focus from mechanical tasks to creativity. Cardoso questioned if contemporary Generative design replaces conventional approaches entirely. While acknowledging decades-old technology, he emphasized recent advancements facilitating practical applications. He raised concerns about AI’s future role, questioning autonomous reference citing and ethical remixing practices. Cardoso urged grounding ideas in reality for meaningful progress. He discussed data as material, hybridity in design, and meaningful human-machine interfaces. Transitioning to construction, he introduced ReAC, revolutionizing construction processes with robotics. Emphasizing critical reflection, he addressed ethical considerations and the balance between technology and human creativity. Cardoso’s lecture highlighted the importance of understanding the limits of architecture within the context of Generative design. While technology opens new possibilities, it’s essential to recognize the constraints imposed by materiality, cultural context, and environmental sustainability. Some of the points mentioned by Professor Cardoso, that caught my attention was his statement that we should see the existing problems in their real size “There are some difficult problems we shouldn’t try to avoid or simplify, because in reality they’re difficult problems”.
In his lecture professor Llach explores the contemporary theme of AI in architecture. Today there are new experimental technologies to generate computational design. The goal is to go beyond the limits of architecture and overcome them by solving problems posed by architectural practice with these new tools. Infact, AI can use a lot of combinations of technologies to create an innovative computational approach to architecture. For example, AI can automatically transfer knowledge about a specific project with a digital workflow (composition of framework) that can be used for others works. AI can develop a complex geometry, generate a sort of data-drawing design, and organize system of information to choose data that can be use. AI can collect ideas, data, images, can share and combine these based on specific characteristic (data or geometry or aspect). But all these tools are technologies that have their specific use and present some limitations. In fact, AI doesn’t change radically problems of architecture. The goal will be to drive this technology and measure what we have with what we need, essentially as something efficient and simple.
Prof. Cardoso Llach proved to us that AI is not a new thing at all. He introduced AI applications reasoning develop way back in the 1950’s.
According to him the only reason those developments didn’t arrive the AI level of today is because back then there was not the computer power.
we have today. Examples of research done included one where music notes were given to check their validity against stylistic musicians’ “rules” that were created.
A similar approach was tested in architecture. Research tried to create a “grammar” of the Architect Paladio and use it to develop similar buildings.
AI allows us to look at things from distance, to create robotic solutions while visiting the company first, understanding their needs. The “interface” allows us to create better ways for visualizing data and by that recognising pattern. Pattern clustering and correlation learning between structure and traits could improve urban planning.
L’argomento trattato dal professore è di molta attualità; l’intelligenza artificiale; certamente molti di noi, non abbiamo ancora avuto modo di conoscerla, ma, come un po’ tutte le cose che non si conoscono, ingenera in tutti noi un atavico timore: quello che la “macchina “possa prendere il sopravvento sull’uomo, uno scenario che trae origini dai tanti libri e film di fantascienza.
Il professore Cardoso, ci ha comunque dimostrato che l’intelligenza artificiale non è affatto una novità, difatti già negli anni ‘50 con il lavoro di Brooks e dei suoi collaboratori avevano sviluppato un metodo geniale che consisteva nel generare dei brani musicali attraverso una sequenza numerica.
Sempre secondo il relatore lo sviluppo di livello di applicazione dell’intelligenza artificiale va di pari passo con l’innovazione informatica.
L’impiego dell’intelligenza artificiale nella progettazione urbana e in altri campi creativi interessarono da subito gli architetti, che svilupparono un metodo chiamato “grammatica della forma” un metodo di progettazione fatto di introduzione di informazioni e di formule matematiche che governano il modo in cui in cui le forme possono essere combinate e trasformate.
La discussione sull’integrazione dell’intelligenza artificiale nell’architettura può essere collegata al tema generale della conferenza “i limiti dell’architettura”, poiché sfida i confini tradizionali evidenziando allo stesso tempo la necessità di un’applicazione responsabile e consapevole.
In the discussion by Prof. Daniel Cardoso Llac of the broad historical progression of artificial intelligence (AI), he makes clear that the name “AI” encompasses a wide range of instruments and that it always conjures up images of something sinister and larger than life.
Although professor Cardoso Llac emphasizes that the concept of automated machines and automatic design has been around for a while, AI is already being used to assist in some architecture projects. It was a very educational talk that began with an introduction to AI-based generative design in prehistory and then covered its modern applications. By using these new technologies to solve difficulties given by architectural practice, the aim is to transcend the boundaries of architecture and conquer it. In conclusion, his talk emphasized the necessity of implementing AI with consideration and sustainability while also outlining the possible advantages and difficulties of doing so in the field of architecture. The conference’s overarching theme, “the limits of architecture,” can be connected to Llach’s discussion on AI integration in architecture because it pushes conventional bounds and emphasizes the significance of moral and sustainable behavior.
Professor Daniel Cardoso Lach showed us in his seminar his point of view on the theme of AI. He starts from the historical examples of the late ‘900 that testify as some technologies that today are very prominent, already existed years ago but had fewer applications than today. The presence of man was fundamental once while today with new technologies and the increased computing power of AI, the mechanical tasks to be performed and the creative part of these works seem to be more divided than before. To date it is also possible to think of several new directions for the use of these intelligences that are based mainly on data: use them in a hybrid way to change maybe even our way of thinking or as a tool to change the quality of the spaces that we architects design. The limits with these technologies may seem few but also many and binding in some aspects, but certainly for now this technology still immature for what you ask an architect but with large improvement for the future.
I found Daniel Cordoso Llach’s lecture very difficult to understand many points are unclear to me. What I did understand is that in the course of the lecture, he presents us with an overview of the evolution in the use of artificial intelligence. It starts from a use in the field of music to applications related to architecture.
As in the case of the other architect Fredy Fortich, who also dealt with the subject of artificial intelligence, he also points out that it is still a tool, an additional tool, but one that cannot perform any real function on its own.
The aspect that touched me most was when he dealt with this tool’s ethical and environmental limits. This topic was dealt with by taking as an example various published articles comparing different uses of artificial intelligence and the energy equivalent needed to perform different types of tasks. Personally, I did not think they required so much energy.
Daniel Cardoso Llach talks about the impacts of Artificial Intelligence on architecture. MVRDV uses AI and other computer driven methods in various projects. Such as the Dutch Expo Pavilion and the Market Hall in Rotterdam. MVRDV also uses AI in brainstorming, reference generation, conceptualization, and CAD design. They use programs like Rhino, Revit, Grasshopper and Dynamo .Daniel Cardoso Llach adamantly stated that AI is a tool and not an architect and discuss the differences between AI and deep learning, artificial intelligence. Daniel Cardoso Llach also talks about the use of AI in creating collages, landscaping, rendering and upscaling images using stable diffusion. Despite the limitations of AI, Daniel Cardoso Llach believes it can contribute to architectural design as a tool but it can not replace human creativity. Daniel Cardoso Llach also emphasizes the importance of learning to modify and direct the AI to a more defined end goal oriented version.
In his lecture, Prof. Cardoso Llach delves into the intersection of technological innovation and cultural perspectives in architecture and design. Highlighting the enduring fascination with autonomous design machines among architects, designers, mathematicians, scientists, and engineers, he examines how these ambitions reflect evolving understandings of design. With the advent of data-intensive “generative” computation, the landscape has shifted dramatically. Prof. Cardoso Llach explores the historical roots of generative design methods and examines the implications of integrating machine learning techniques into architectural practice. Through his discussion, he offers insights into the complex interplay between technology and culture in shaping design processes. Drawing from recent research at Carnegie Mellon University’s Computational Design Laboratory, he outlines new directions for leveraging these advancements in design, architecture, and construction. This lecture promises to provoke thought and inspire innovation in the field.
Professor Cardoso’s lecture on Generative design explored its history, concepts, and future directions, emphasizing its role in fostering diverse viewpoints and creativity. He discussed the integration of computer capabilities in design, highlighting recent advancements while questioning the potential replacement of conventional methods. Cardoso also raised concerns about AI’s evolving role and ethical considerations, emphasizing the importance of grounding ideas in reality. Transitioning to construction, he introduced ReAC as a robotics-driven innovation. Fredy Fortich complemented these insights with MVRDV studio’s methodologies, focusing on parametric design and AI. Fortich highlighted AI’s potential in city organization and various project phases but stressed the need for human oversight. He initiated debates on coherence, sustainability, and copyright issues in AI-driven design. Both presentations underscored the complex interplay between technology and human creativity in design innovation.
The lecture by Prof. Cardoso Llach was very captivating and interesting. Before I did not know where the roots of this new technology were originating to. It was very suprising to see that the AI that we know and started to use very recently was not a recent tool at all, but it was dating back to way before. The history of generative design was in the process all along since the generation of computers. It was interesting to see the examples and works he is working on in the lab, the musical one also creating a pattern. But of course, all the good things aside, he stresses that there are always ethical questions rising as well. Such as the copyrights etc. Overall, AI is an interest in many fields, from architecture to urban planning, to programmers to musicians.
Daniel Cardoso Llach’s lecture on “Frontiers of Artificial Intelligence in Architecture” takes a deep dive into the history and future trajectories of generative design methods in architecture, particularly focusing on the implications of integrating artificial intelligence and machine learning techniques into this field. The recent adoption of AI and machine learning in architecture represents a significant shift in how design processes are viewed and implemented. Cardoso Llach discussed both the potential and the pitfalls of these technologies in architectural contexts, addressing issues like creativity, authorship, and the role of the architect in an AI-driven design process. His insights helped pave the way for thoughtful integration of these technologies in architectural practices, ensuring that they enhance rather than overshadow the creative and social imperatives of architecture.
Fredy Fortich’s lecture on how MVRDV is integrating Artificial Intelligence into its architectural design processes presents an exciting overview of contemporary practices at the intersection of AI and architecture. MVRDV, known for its innovative and often unconventional architectural solutions, appears to be pushing the boundaries of how AI can be incorporated into every stage of the design process. Fortich started by discussing the history of MVRDV, emphasizing its roots as a practice that leverages data extensively to inform its design decisions. This historical perspective could provide insights into how the firm’s approach to data-driven design has evolved and set the stage for the adoption of AI. Fortich’s lecture provided a comprehensive look at how AI can not only streamline and enhance the design process but also potentially transform the way architectural practices operate, leading to innovations that were previously unimaginable.
The “Frontiers of Artificial Intelligence in Architecture” lecture by Daniel Cardoso Llach examined the development of generative design techniques in architecture, emphasizing on the implications of incorporating machine learning and artificial intelligence methods. The recent use of AI and machine learning in architecture signifies a dramatic change in the way that design processes are seen and carried out. In his discussion of these technologies’ possibilities and drawbacks in the context of architecture, Cardoso Llach touched on topics such as authorship, creativity, and the architect’s place in an AI-driven design process. His understanding of these, helps new technologies to be carefully incorporated into architectural methods, ensuring that they complement the numerous aspects of architecture than compromise on any major ones.
Fredy Fortich provided an overview of current practices at the nexus of AI and architecture, discussing how MVRDV is incorporating AI into architectural design processes. MVRDV is known for its creative and unorthodox architectural solutions and is at the forefront of applying AI in the design phase.
Fortich delved into the origins of MVRDV, emphasizing its foundation as a data-driven methodology that informs design decisions and sets the stage for the integration of AI.