AI and Generative Design

(Image attribution: Cryteria, CC BY 3.0 <https://creativecommons.org/licenses/by/3.0>, via Wikimedia Commons)

In their most recent Life of an Architect podcast, Andrew Hawkins and Bob Borson discussed how the development of artificial intelligence is intersecting with both the practice of architecture and the education process for students in schools of architecture. For this episode, Andrew and Bob brought in Kory Bieg, Associate Professor of Architecture at the University of Texas at Austin, and founder of OTA+, an architecture, design, and research, an office that specializes in the development and use of current, new, and emerging digital technologies in design and construction. Andrew, Bob, and Kory tackled the “Everest-sized” task of discussing the Wild West reality of AI and its increasing accessibility. How AI will increasingly impact the way we design and think about architecture is certain to be profound.
 
Computers are great at sorting tons of data. Machine-learning models can analyze patterns humans have trained them to recognize within massive data sets. Andrew, Bob, and Kory speculated that extrapolation beyond text-to-image platforms, such as Midjourney or Dall-E, may lead to the translation of imagery to actual building projects. As they noted, such a capability may be down the road a bit, but it is coming quickly. Despite AI’s potential for destabilizing and disruptive change, the three concluded it will not completely transform the architectural profession. Optimistically, Kory believes machine platforms will never entirely replace humans in the design of buildings, precisely because they are not human. The three concluded there is plenty to be positive about this innovative technology, and I agree.
 
One way it is conceivable artificial general intelligence (AGI) could autonomously design a building is through a process called generative design. Generative design uses algorithms and AI to generate multiple design options based on a set of input parameters, such as site constraints, functional requirements, and material properties. The AI can analyze and learn from a vast amount of data on building design, construction techniques, and environmental factors to optimize the design process and generate solutions that meet specific goals and constraints. If programmed with the necessary knowledge of architectural design principles, building codes, and construction methods, an AGI system could analyze countless options and use optimization algorithms to refine a project based on specific criteria, such as cost, energy efficiency, or aesthetic appeal. The system could consider real-time data, such as local weather or energy usage, to further refine the design.
 
Currently, designers are already using AI to automate complex, data-driven tasks such as the analysis of a building’s probable energy performance. These applications are likely to expand and improve over the next decade, leading to greater automation and optimization in the building design process. However, for the near future human input will remain necessary and critical to the production of real architecture. My prediction is AI tools will soon become powerful enough that architects can simply use natural speech or text as prompts to communicate with a comprehensive AI in the process of generating successive design iterations. These iterations would reflect the inclusion, integration, and synthesis of many essential design concerns. Human architects would successively refine the output by providing the AI with feedback for each design iteration until the design reaches a point they (the humans) deem optimal. As Kory pointed out in the podcast, how people analyze the iterations will prove more important than their initial prompts.
 
Some people fear the advent of AGI and the Singularity, that hypothetical moment in the future when an AI system becomes capable of recursive self-improvement. They are afraid the development of AGI may have significant and unpredictable consequences for humanity. Recent news headlines about a Google software engineer claiming that the company’s LaMDA chatbot had become sentient have fueled this fear. The proliferation of other rapidly improving language models such as ChatGPT additionally has furthered the widespread belief the Singularity is imminent. And of course, there is the hackneyed science fiction trope about artificial intelligence going haywire and killing the human species to dominate the world.
 
Tech companies are promising to use AI to revolutionize how we do everything, including architecture. But for now, much of the current AI hype is exactly just that—hype. AGI capable of fully autonomous building design is still a long way off. Achieving true AGI will require significant breakthroughs in AI research and development, as well as dramatic advances in computing power and data processing. Notwithstanding what some may believe, this will continue to be a gradual process that occurs over many years. Rest easy—the robots are not coming for us anytime soon. The AI needs us and our humanity for input.
 
If not yet fully AGI, rudimentary artificial intelligence is nevertheless here. It is improving with exponential speed. AI promises to make the future work of architects better and easier so we can focus on what fundamentally distinguishes real architecture from mere building.