“AI - Architectural Intelligence” at ZHAW

Digital Aids with Risks and Side Effects

Elias Baumgarten
14. June 2024
Photo: Elias Baumgarten

It is not artificial intelligence (AI) that is dangerous, said Elena Gavagnin, but the people who use it. The naivety with which some contemporaries use AI applications, such as the currently ubiquitous ChatGPT, makes you want to agree with the expert. But does her assessment also apply to the architecture? Gavagnin researches and teaches at the Institute of Information Systems at the ZHAW. She came to the “AI – Architectural Intelligence” conference to explain how AI actually works. Her core message: AI is not an omniscient counterpart that knows the answer to every question. Rather, it is based on probability calculation. A chatbot like ChatGPT only gives the most likely correct answer. And even more so, such programs cannot develop creative solutions or devise clever strategies. In short, according to Gavagnin and other experts, "intelligent" is actually the wrong word for such systems. The term invites fallacies.

But what is AI particularly good at? Bige Tunçer, who teaches at Eindhoven University of Technology, counts sifting through large amounts of data for relevant information and recognizing complex patterns among its greatest strengths. AI is also superior to humans in making predictions based on existing data and in optimization processes. However, it lacks common sense and empathy as well as intuition and the ability to transfer its conclusions to other areas. AI also performs poorly when it comes to understanding context and causalities or recognizing when reframing is necessary — all skills that characterize good architects.

Those who produce images do not yet design

So AI is not an issue in architecture? Not at all. There are already plenty of applications — from design helpers to tools for analysis and optimization. The programs DALL-E (2021) and Midjourney (2022) can generate images based on text input. Alexandre Theriot had his students work with them in his design course at ETH Zurich. The visual worlds were very inspiring, he reported in his lecture, but nothing more: the students soon complained that it was almost impossible to control the image production. Nevertheless, tools such as DALL-E and Midjourney are likely to become part of the design process for many architects in the near future — or are already part of it today. And there is nothing wrong with that. As Georg Vrachliotis said in Winterthur, AI is a new tool. This does not mean that sketching pencils and model-making tools have become obsolete.

However, it becomes dangerous when two-dimensional images are mistaken for architecture. This is because the diverse sensory qualities of a building are not taken into account, nor is the context. “Designed” in this way, architecture would only consist of shapes and surfaces; buildings would remain object-like and isolated. So if you want to benefit from AI when designing, you need excellent spatial knowledge and imagination. But this is precisely where things are going wrong: the focus on images threatens to stunt spatial perception. Patric Furrer, Andreas Jud and Stefan Kurath lament the consequences in their book Digitalization and Architecture in Teaching and Practice today. They write that architecture is increasingly being created that only appeals to the eyes while neglecting other sensory impressions. This is where architecture schools need to take countermeasures. The more that work is done in digital worlds, the more important physical models become: they train spatial perception and allow students to check the sensory qualities of their designs. It is no coincidence that models and drawings have always been important design tools for the pioneers of digital architecture, such as Frank O. Gehry and Zaha Hadid. This would probably also be entirely in the spirit of mathematician and philosopher Dieter Mersch, who warned us at the ZHAW that AI could have a negative impact on our way of thinking. He is also convinced that, due to its strictly mathematical basis, it will never produce art, but merely imitate it. This is because art defies logic and includes the random.

Image: ZHAW, via espazium
The idea is just the starting point

“The work of architects does not end with the idea,” ZHAW professor Stefan Kurath said during the closing discussion, “it is just the beginning.” Architecture is created in a complex, often lengthy negotiation process: architects have to take countless social, economic, political, and ecological factors into account when planning and ultimately implementing a project. They need great social skills and have to negotiate wisely and forge alliances. In times of climate change, increasing scarcity of resources, rising social tensions, and simultaneous processes of growth and shrinkage, this task becomes even trickier. AI, which can only reproduce what already exists, is not capable of mastering it. Human skills are still required here, and in-depth knowledge of the link between architecture and society remains essential.

How to make use of it?

In line with the conference title, “AI – Architectural Intelligence,” good projects will probably be created in the future through the interaction of artificial and human intelligence. It is important that we architects — indeed all people — develop a precise understanding of AI and that we are clear about where its strengths and weaknesses lie. Instead of anxiously asking whether AI will take our jobs, we should be interested in how we can use it to build in an ecologically and socially sustainable way. We need to stay on the ball, because AI is currently developing at a rapid pace. And that is why this highly interesting event at the ZHAW, organized with great dedication by Andri Gerber and Michael Mieskes, should have been better attended. After all, anyone who shies away from the topic because it arouses fears or is being discussed too frequently is leaving the field to others and possibly missing the opportunity to play a key role in the construction process of the future. This can have unpleasant consequences: In Elena Gavagnin's words, none of the programs discussed in the context of “AI – Architectural Intelligence” are a danger. The situation is different if they encounter a superficial understanding of architecture, a lack of awareness of the interaction between the built environment and society, or insufficient spatial knowledge. AI is no substitute for expertise.

This article was originally published as “Hilfsmittel mit Risiken und Nebenwirkungen” as Swiss-Architects. English translation edited by John Hill.

Other articles in this category