Computer-aided architectural design has existed for quite a while now and with the proliferation of parametric design tools like Grasshopper, that potential can be projected to hit new heights. More recently, architects like Stephen Coorlas has explored the potential of artificial intelligence-aided architecture using Midjourney, a text-to-image AI process, to create renderings and concept images.
The concept behind using AI for architecture seems very straightforward when you think closely about it. Just like art, architecture is, for the most part, still represented through 2D imagery. Conceptual sketches are often how most designs start. But with the aid of AI processes like Midjourney and DALL-E, a simple text prompt with stylistic keywords can achieve more creative and faithful rendition of those initial ideas in a fraction of the time.
And once the architect receives these initial images, they can proceed to tweak and adjust their designs according to them or abandon those ideas entirely. In that sense, the use of AI for the ideation process of architecture is already quite powerful. However, it doesn’t just stop there. For the most part, the labor-intensive portion of architecture is adjusting layouts and drawing construction details. With the help of AI, these processes can be streamlined and simplified. Companies like Smartscapes Studio have already mapped the idea of using AI to create responsive designs and it won’t be long before construction drawings become an entirely automated process.
I believe there are two parts to this conversation: using AI to create imagery that inspire architects’ initial designs and automating labor-intensive tasks during the construction documentation phase. As of right now, the architect’s role still remains relevant as they serve as the intermediary between turning the abstract concept into an actual building. However, once we figure out how to bridge that gap the role of the architect will be diminished even further.
In terms of how the AI’s data set is gathered, Midjourney currently uses a proprietary AI program and therefore not much is known about their processes behind the scenes. But much like other open-sourced AI programs like DALL-E, they probably use a combination of real data and synthetic data to train their AI, albeit their output seem to be more aesthetic when compared to DALL-E.
Of course, any talk of AI-generated imagery comes with its ethical implications. In short, I believe the latter part of the application of AI in architectural construction lends itself to less controversy since it is simply an automation of the work that would normally be done by a human. And since many construction details end up being similar if not identical, I am surprised this process is not more widely automated already. As for the use of AI in initial ideation of design concepts, an argument can be made for how it takes away from the creative processes that architects normally undergo and in turn, stripping them of that agency. But the same can be said for the role of precedents I believe. When architects design, they are rarely starting from scratch. There is always tension between what has already been made and what new elements or characteristics the designer can embellish. Therefore, it is a tricky territory to navigate and as more and more architects begin to discover the capabilities of AI in conceptual generation, that reality will become more transparent.