Similar to last year when I attended and wrote about a World Architecture 100 (WA100) Live webinar focused on technology (see “World Architecture 100 Webinar: Technology for Architectural Innovation”), I attended a similar webinar this year which discussed how AI (Artificial Intelligence) was impacting the architecture sector. While it is still relatively early days for AI in design (see my 2022 article, “AI in AEC Updates, 2022”), it was very interesting and insightful to hear the perspectives of some of the leading firms in the UK on AI technology and learn about the exploratory AI work that they have started working on. This article captures the main highlights that stood out to me from the webinar.
Foster + Partners is not only one of the leading architectural firms in the world (Figure 1) but also one of the most advanced in its implementation of AEC technology. (See the article, “Technology at Foster + Partners” authored in 2020 by Han Shi, Head of BIM & Design Systems at the firm as well as the article on last year’s WA100 Live event in which the firm participated.) This year, we had Martha Tsigkari, senior partner and head of the Applied R+D (ARD) group at Foster + Partners (Figure 2), return to share AI-specific developments at the firm.
The approach to AI at Foster + Partners is to use it to augment its designers’ creativity and problem solving capability. For this, the key is to recognize that in order to harness the power of AI, we first need to be able to control what powers it, which is data. One example of this is an internally developed tool called Hydra for urban design that produces a solution space of hundreds of thousands of options and their performance — with over 12 different types of performance analyses — within a matter of hours. These automatically generated solutions satisfying specific performance criteria are used as the dataset to train the AI engine, which is then used to start predicting different configurations for the designer to consider for a new urban design project (Figure 3).
Other examples of the use of AI include replacing slower analytic processes with much faster predictive ones that may be 95% to 98% accurate instead of 100% but can provide analysis results within seconds rather than hours; creating knowledge bases of company resources comprising millions of documents that can be searched using natural language, not just for textual information but also for designs, models, and details (Figure 4); and deriving operational insights from business data to understand trends and optimize performance and processes for the firm.
AI is also being used in the very specific area of material science to understand and predict how passively actuated materials would react to temperature changes, which in turn would help to specify how materials should be layered for desired deformations. Another compelling example was the internal development and use of “stable diffusion” — now a key AI technology for creating photorealistic images from text prompts that are the mainstay of popular AI tools like Midjourney — to assist designers with ideas directly within the design tools they are using like Rhino (Figure 5). In addition to these ideas coming directly from the firm’s own designs, the computations themselves are being run through their own servers, so the IP and copyright stays with Foster + Partners, which is not the case when external tools such as Midjourney are used.
And finally, Foster + Partners is developing a collaborative Machine Learning app which uses its stable diffusion technology to power a platform that multiple people in the office can contribute to. They can each undertake their own individual explorations with the technology, running their own prompts, and share their results with others at the firm to spark creative design ideas for everyone.
The potential of AI in architecture from the perspective of a smaller architectural practice was presented by Nathalie Rozencwajg, the Founder and Director of NAME Architecture, which works in the cultural, residential, retail and hospitality sectors. Most of its projects so far are in London and Paris where it is based, but it also has some international projects that it is working on (Figure 7).
NAME Architecture is exploring the use of AI in augmenting design along several fronts, including conceptualization, planning and analysis, visualization, sustainable design, and client engagement. For conceptual design, AI tools like Midjourney are being used to generate original and unique references for testing design aspects such as materials, lighting, atmosphere, etc., to inspire and guide the creative direction of a project. A more advanced use of AI at NAME Architecture for conceptualization is custom-training AI tools like krea.ai and ChatGPT Builder to generate images and text in alignment with their own distinctive style of design. Both these uses of AI are illustrated in Figure 8.
For planning and analysis, NAME Architecture is exploring the use of tools like Autodesk Forma (earlier known as spacemaker.ai) for analyzing data and developing massing options for site planning, as well as tools like Planfinder.ai and Maket.ai for generating layout options based on specific criteria (Figure 9). For visualization, it is testing a whole slew of tools: Lookx, Midjourney, and Prome.ai, for quick renders; runway and Pika for animations and enhancing presentations; and Adobe Firefly for post-production work to add details to renderings (Figure 10). AI tools are also being tested to explore different aspects of energy-efficient design as well as for creating mood-boards and enhanced presentations for client engagement.
Heatherwick Studio is a 250 people strong design studio based in London focusing on large scale projects in cities all over the world. It has won over a hundred international awards for design excellence, including the Prix Versailles and the RIBA National Awards. A snapshot of its design work, as presented at the WA100 Live webinar, is shown in Figure 11.
Heatherwick Studio’s use of AI and ML was presented by Pablo Zamorano, its Head of Geometry and Computational Design. In addition to describing some of the work that is being done using AI tools, the focus was on discussing which ones can be most useful. Similar to other architectural firms, Heatherwick Studio started with exploring tools like Midjourney for visualization (Figure 12), but while the initial reaction to Midjourney was that it was almost magical, after some time it was found to be quite generic and somewhat boring. There wasn’t much differentiation between the visualizations that were produced even if the inputs that were provided to the tool were varied.
The disappointing results with tools like Midjourney made the computational team at Heatherwick Studio decide to find ways to customize AI so that it could be used in a way that reflected the unique design and visualization style of the firm. This was done by training the AI engine with the large number of models created over the firm’s 30 year history (Figure 13), resulting in the AI tool being able to generate design options and visualizations in the specific style of the firm, making them much more useful. A couple of examples are shown in Figure 14.
In another especially intriguing example, a massing model image is divided into two parts, with each part using a different set of references, as shown in Figure 15. The two parts are then consolidated together in one rendering. Here again, the references come from Heatherwick Studio’s own library of designs, so the designers already have a head start when it comes to conceptualizing the project.
And finally, it was important that the designers should be able to use these customized AI tools within their design application, and this was done by developing a plug-in for the tools to Rhino (Figure 16).
With regard to AI in architecture, the most urgent question seems to be whether AI will replace architects. This is a question that is actually being asked across all industries, especially in creative fields such as art, writing, music, etc. I think that this fear is overblown as, to put it very simply, AI works by learning from past creations and therefore cannot create brand-new ones. (For example, I doubt that AI could “attend” the WA100 Live webinar and write this article, but now that it is written and available in the public domain, AI can use it as one its learning resources to write knowledgeably about AI in Architecture!)
Also, as Martha Tsigkari of Foster + Partners pointed out, we need to keep in mind that our fear about technology encroachment is not new – it has happened before and will happen again. Examples include the invention of the printing press for society as a whole and the trepidation with which it was viewed, the introduction of CAD in the architecture industry, or the splitting of the atom, which was viewed as a catastrophic event for all of humanity.
But we are still here and still very much creating!
Lachmi Khemlani is founder and editor of AECbytes. She has a Ph.D. in Architecture from UC Berkeley, specializing in intelligent building modeling, and consults and writes on AEC technology. She can be reached at lachmi@aecbytes.com.
Have comments or feedback on this article? Visit its AECbytes blog posting to share them with other readers or see what others have to say.
AECbytes content should not be reproduced on any other website, blog, print publication, or newsletter without permission.
This article takes a closer look at the innovative tools for design automation being developed by the A/E firm, Bryden Wood, including applications for school design, housing, and motorways, as well as its “factory on site” approach to industrialized construction.
This article explores the application of AI (Artificial Intelligence) in tools across each of the three main processes in AEC: Design, Construction, and Operations/FM. Some of the applications discussed are TestFit, Reconstruct, and ALICE Technologies.
Han Shi, Head of BIM & Design Systems at Foster + Partners, describes how technology forms an integral part of the firm’s workflow, with several interdisciplinary groups involved in computational design, building physics, performance analysis, optimisation, fabrication, and interaction design.
Augmenta is a new AEC technology solution being developed that is using the breakthrough technology of generative AI to automatically create detailed, code-compliant, and constructible designs for different engineering disciplines in Revit.
