ARTIFICIAL INTELLIGENCE IN ARCHITECTURE: ZAHA HADID ARCHITECTS' PIONEERING USE AND COLLABORATION WITH NVIDIA

By Jonathan Bell

Published on: December 13, 2024

Source: Wallpaper*

The debate around generative AI in architecture and creativity in general seems set to continue for a long time. Creatives are divided: some see AI as just another tool, while others view it as an existential threat that could unravel or even destroy the “magic” of human creativity.

Although architecture considers itself the “Mother of the Arts,” it is actually a vast field where many disciplines and complex data intertwine. So, can AI tools offer a way to solve the complexity of modern construction without compromising the visual expression of architectural design?

Founded in 1980 by the late Zaha Hadid, Zaha Hadid Architects (ZHA) has always been a pioneering studio in computer-aided design. Wallpaper* spoke with Shajay Bhooshan, founding partner and Associate Director of the studio’s Computational Design Research Group (CODE), Lead Designer Vishu Bhooshan from the same team, and ZHA Director Nils Fischer, who has been with the firm for 20 years. How have machine learning, AI, and CAD (computer-aided design) evolved and intersected over the years?

AI in Architecture: What Does the Future Look Like?

“We have been using machine learning for floor plan optimization for many years,” says Fischer. “Many people call that AI.”

The studio quickly began exploring the potential of early generative software (such as OpenAI’s DALL·E introduced in 2021). Fischer says, “We were as excited as anyone,” adding, “We’re not afraid to break things.”

With 1.4 million followers on Instagram, ZHA is the world’s most popular architecture studio and has a huge visual archive—both public and private. One of the studio’s first AI experiments involved uploading this dataset into a generative AI model called Stable Diffusion.

Fischer explains:

“Our team curated the datasets and trained the models. But these models are like five-year-old children—they can dream wonderfully, but they can’t yet produce what you actually need.”

Text-to-image AI models were trained to generate structurally and contextually meaningful results. In the process, many “fantasy” suggestions emerged that enriched the design language but also required serious curation (selection, filtering).