For a culture of sharing

What hopes do architects have for artificial intelligence? For our 1/2.2025 issue, we asked around 20 experts. Thomas Wortmann, from the Institute for Computational Design and Construction at the University of Stuttgart, hopes that the triumph of AI will be accompanied by a stronger open-source culture in architecture firms.

Where do you already use artificial intelligence (AI) in your work and what has been your experience with it?
My department at the ICD regularly uses and develops AI methods in research and teaching. We mostly use self-trained AI models instead of existing models like Midjourney or ChatGPT. For example, we use machine learning to predict values in real time that either require time-consuming simulations or for which no simulation models exist. Examples include energy demand, daylight quality and wind loads in buildings. Other examples are the behaviour of bio-based inhomogeneous materials such as wood, and manufacturing times in robotic assembly. These AI models allow us to support design decisions with real-time information and accelerate optimisation processes.

In addition to these data-based AI approaches, we also use a rule-based, symbolic approach, known as knowledge graphs, for a more flexible way of modelling with BIM. This approach allows any concept to be represented in a standardised way, data integrity to be checked, and reasoning about these concepts to be performed automatically. This enables different disciplines in the building design process to exchange data without loss, for example to check compliance with fire regulations. Finally, we also use graphs to generate three-dimensional building designs and BIM models with meaningful spatial relationships using graph-based neural networks.

What are the limitations and risks of AI?
The main limitation of AI is that while it learns patterns from data, it has no understanding of our world. AI does not understand basic spatial concepts such as up and down, left and right, or physical concepts such as gravity. If you look closely, this is easy to see in AI-generated images and videos. More generally, today's AI models such as ChatGPT are not reliable and often give incorrect information, for example about building rules. The biggest risk is to put too much trust in the results of AI models.

A related limitation is the lack of data on buildings, architecture and construction in general. The fact that AI models only learn from data, and the amount of data on buildings, architecture and construction on the internet is limited, is one reason why these models are not very reliable in this area. Currently, more (electrical and mental) energy is being put into video processing. Processing large amounts of video could improve spatial understanding and the reliability of AI models in general. Another approach to improving the reliability of AI models is to combine data-based AI methods with symbolic AI methods, known as neuro-symbolic AI.

How will AI change the design and construction process over the next decade?
Symbolic, or rule-based, AI can be used to automate many repetitive design tasks, such as residential and office buildings. In addition, of course, image-generating AI models can speed up visualisation tasks enormously. I therefore expect architects to focus on specific and culturally relevant design tasks in existing buildings.

What developments in AI do you hope to see in the future?
I hope that AI will become more reliable and increasingly develop spatial understanding. This is something we are working on in my department. Another important step in this direction would be for an open data culture of collecting and sharing data to become established in architecture and construction, analogous to the open source culture in computer science.



All interviews on the topic of AI can be found in our white paper on Architecture + Artificial Intelligence.

Read more in Detail 1/2.2025 and in our databank Detail Inspiration.