AI in Architecture: A Creativity Boost in Design with Images
What does Artificial Intelligence offer us? In this interview, digitalization expert Stefan Kaufmann answers five of the most pressing questions about the future of AI in architecture.
As the person responsible for BIM strategies and new technologies in product management at ALLPLAN, Stefan Kaufmann always has his finger on the pulse of technological developments. In this interview, the AECO digitalization expert answers five questions about the role of artificial intelligence in architecture.
Mr. Kaufmann, how can AI support architects in their work?
The use of artificial intelligence is very diverse. Often, we won’t even notice that a neural network is working for us in the background. Every software solution will have a chatbot that increasingly supports us intuitively and automatically, making our daily work better. AI assistants will help us detect mistakes in our work. Some tasks will be completely taken over by AI, making future BIM projects easier: there will be fewer rules and more AI. AI systems can also support the joint use of differently structured BIM models.
The management and use of different file formats from different systems and versions is currently a significant challenge. AI will play a role here. When AI learns to interpret building data, there will be fewer data silos. It will also structure "dumb" data retroactively. For example, we could extract BIM models from a simple point cloud or automatically retrieve information about a building's energy consumption or structural behavior from 2D plans. In the future, I expect AI to assist us in using data sources for applications they were not originally created for.
How can artificial intelligence improve the efficiency of architecture and planning offices?
Access to up-to-date specialist knowledge is a challenge for architects. They face the difficulty of designing under increasingly complex conditions. A great deal of expertise is needed, not only in highly complex hospital construction, for example, but also in designing single-family homes. It is therefore increasingly important to have efficient and intuitive access to knowledge to ensure that the design remains technically and regulatory feasible. AI will enable us to receive feedback earlier during planning, avoiding costly late-stage corrections.
In the near future, we expect many AI assistants to take over tedious routine tasks in the planning process. For example, if AI recognizes that I am currently designing a window, why shouldn't it assist me in designing the next twenty windows? However, we must also move away from associating AI exclusively with 3D planning. It is quite possible that classic 2D-based planning processes will become more attractive in the future, as they may be better supported by AI than complex 3D-based planning processes. Even outside of constructive planning, AI can help architects, for example, with communication in international teams.
Can AI generate 3D models?
Yes, it is possible with very simple geometries – anyone can try it. You can generate geometry with voice commands (prompts) or ask for suggestions to improve a static system. These are early approaches; however, the question is how useful the result and process are. Large language models are very powerful, especially when it comes to generating text-based content, and geometry can be described in many ways textually. However, you won’t get a reliable and precise model just from prompts. With CAD software, you can execute your design idea to the nearest tenth of a millimeter. We are still far from that with language models. Therefore, I believe in a combination of traditional parameterized modeling approaches and machine learning for now.
What AI products can offices expect from the market and ALLPLAN in the future?
We are currently working in four different areas: assistants, automations, optimizers, and data managers. There will be chatbots that become more intelligent and can answer questions on a wide range of topics. These assistants will initially respond to requests and may later understand the context in which you ask a question.
Automations will know what you are currently doing. AI will understand and evaluate the design and be able to predict the next steps. In the early phases of design, we are already seeing image synthesis models being used. At the same time, we are further developing our parametric design tools, including what is known as symbolic AI.
Optimizers analyze the planning and construction outcomes and check for quality, safety, performance, and compliance with regulations. For example, we have invested in the startup Imerso, whose solution monitors construction quality on-site using AI to compare point clouds with planning models.
Data managers improve the usability of data. Today, when ten people collaborate on a BIM project, they must coordinate closely to ensure that the models can be used cleanly at the end. This never works 100% and leads to frustration and extra effort. Yet AI can translate between languages and synonyms, even within a BIM model. This could significantly simplify the work of BIM managers in the future.
What does the future look like? Will we end up with large AI models that accompany the entire building planning process?
There are different trends, of which one is around large foundational models. One day, such a model might be able to represent all construction knowledge for buildings. However, it is currently too inefficient to call up a large model for every small request. There is a second trend aiming to develop many AI services for specific tasks. Small, fast, and highly specialized AI services are more sustainable because they require less energy for training and operation. The idea is to build workflows from AI-based system components, with the possibility that a single AI could suggest these process sequences.
What can we expect from ALLPLAN in terms of products?
In ALLPLAN 2025, available from October, we have integrated two plugins: Veras from Evolvelab and the AI Visualizer from our own company. These support the architectural design process in various ways. Technically, these are diffusion models trained to generate images. Starting from completely random noise, images gradually emerge that increasingly resemble our inputs. These inputs can be text prompts but also images, such as a screenshot from the CAD system. The results are astonishing. AI creates syntheses based on billions of images and works similarly to a human creative process. Architects and artists also create new things by drawing on existing knowledge and synthesizing it into solutions.
The plugins provide inspiration both in the early idea generation phase of architectural design (e.g., experimenting with different architectural styles) and in later design phases (e.g., representing furniture and materials). The tool is equally suitable for visualizing exterior and interior architecture. It delivers particularly good results when representing materials and plants, as well as creating a specific atmosphere or landscape or urban context.
