Words by Katie Puckett
AI is already supporting architects and engineers — but how far could it go?
Yes, but why stop there?
Rory Hyde / curator of contemporary architecture and urbanism / V&A, London
“It depends how you define the job of an architect. If you really limit the problem to being about materials, engineering and the way that things go together, we’re developing quite complex digital modelling programmes that enable all of those things to be coordinated already. But I would say that working with the client, designing the brief, getting planning permission — that really is the job. I can’t see those being automated any time soon because it is so much about context and people and navigating between the complex demands of different stakeholders.
I think we should be pointing our computers at the bigger questions of the city and housing. The city is a complex system that’s constantly adapting itself through the millions of decisions of individuals but also through the big decisions of policymakers. A machine could try to unpack and play out the consequences of those decisions in real time and perhaps give us a way forward that is more equitable and affordable and fairer. That’s really the promise of AI in architecture. It goes back to Nicholas Negroponte, the founder of the Media Lab at MIT and an architect too, who said that intelligent systems could be more like a partner than a tool, suggesting good ideas rather than just executing the ones that we give them. It sounds a science fiction but things are going along pretty quickly.”
No, it’s just a geeky member of the team
Arjun Kaicker / co-head of Zaha Hadid Analytics & Insights / Zaha Hadid Architects
“When I joined ZHA, I walked in and something seemed different. When I looked across the workplace, all the screens were white, not black as you’d expect with CAD. I wondered if everyone was writing emails. Then I realised that they were scripting and coding. So much design work at ZHA is done computationally. It occurred to me that we could use that same ability to work out what users need from their buildings.
So we’ve created algorithms that evaluate a design based on user requirements. We come up with an option and then run a series of algorithms to assess how well it’s performing on views or daylight, or visibility, or collaboration. Before we would run simple studies manually, but now we can do them a lot quicker in a lot more detail, a lot more objectively. Rather than just comparing two options, we can compare 30. And you can do it within milliseconds.
“The computer never comes up with a design for us, it comes up with a diagram”
The ultimate goal is to have buildings that are self-learning. The design would look at itself and say, ‘these spaces are well used, these spaces are not, people complain when they sit here but they’re happy when they’re somewhere else’. So maybe every three or six months it could come up with its own ideas on how it should change.
The computer never comes up with a design for us, it comes up with a diagram. I think great design will always be about human creativity. This doesn’t take that away, it just augments it. It’s like having a very geeky designer on the team.
We’re using it to find the best location for the core of a building. That’s not a computer designing a building, it’s helping us establish one of the fundamental blocks of a design.
We also used it for view analysis on a conceptual design in Hong Kong. It was really obvious what the good views were, but when we ran this analysis we found that it was different to what we’d expected. A very small change to the angle of the facade, just a few degrees, had a massive impact. It meant that hundreds of people who wouldn’t have had a view before now had a good view.”
It already has… and it looks like this
At ETH Zurich, architect-programmers Michael Hansmeyer and Benjamin Dillenburger have used algorithms to create what is claimed to be “the most complex architectural structure in history”. Digital Grotesque II, built for the Centre Pompidou in Paris, is 3.5m high and has 1.3 billion individual surfaces. The design required 156GB of production data and it was fabricated using a sandstone 3D printer, able to print details to the level of a grain of sand.