San Francisco Chronicle | November 13 by Chase DiFeliciantonio
Can artificial intelligence design a viable building? And, if so, how much could it speed construction of Bay Area homes?
In a housing-starved market like the Bay Area, some property developers are turning to the promise of AI, hoping to cut down on design and building time and save money in the process. But in an area known for its permitting nightmares — especially in San Francisco, where putting up housing takes nearly a year longer than anywhere else in the state — all that computer brain power could find itself wrapped up in red tape.
Artificial intelligence software can speed design and construction. But it’s “not a silver bullet” to fix the lack of cheap homes.
A bevy of cutting-edge products promise to use sophisticated generative AI — think ChatGPT, but in 3D — to automatically run code-compliant wiring through digital buildings, imagine fantastical structures from a sketch, and schedule contractors down to the minute.
One new AI tool from San Francisco software giant Autodesk can spit out thousands of building designs in hours instead of days or weeks, taking into account 10 criteria ranging from cost to carbon footprint to ease of living.
A lot backing up to Interstate 880 southeast of Jack London Square is among the first test cases for this technology jumping from the computer screen to the real world.
Formerly home to the Phoenix Iron Works, the site could eventually host 316 residential units, around a third of them either low-income or supportive housing, as well as thousands of square feet of office and light industrial space.
Working with studio, one-bedroom and two-bedroom units made by Factory_OS, a Vallejo manufacturer of modular housing, designers used the AI software from Autodesk, whose computer-aided design programs are already the industry standard, to rapidly model thousands of configurations of how the buildings might fit onto the lot, and how much carbon would be emitted to get them there.
“Architecture and buildings is about 40% of the global carbon problem,” said David Benjamin, director of architecture, engineering, and design research at Autodesk, whose team applied the software to the project.
“The impossible problem that maybe AI can help with is how are we going to drastically increase the total amount of floor area while we’re drastically decreasing the total carbon emissions from all buildings?” Benjamin said.
Ultimately, two designs — one with a large central green area and the other a more distributed landscape with the smaller buildings sprinkled in a checkerboard arrangement — went before the Oakland Planning Commission. The first, so-called “Central Park” scheme was ultimately chosen, Benjamin said.
Using the software meant “we know exactly what we’re building early on,” said Jamie Hiteshew, director of development at Holliday Development, Factory_OS’ property development arm and the project lead for the Phoenix site. That way, “we’re not churning through a lot of iterations of site plans,” and going back and forth between stakeholders, he said.
Still, property development experts cautioned that AI would solve only part of their problems. Planning and design typically account for just about 5% of total project costs for Bay Area developments, said Carolyn Bookhart, director of real estate development at the nonprofit Resources for Community Development, in an email. Other estimates range as high as 10%.
Bookhart, speaking generally and not specifically about the Phoenix project, said 60% to 70% of project costs typically come from the actual construction.
Designing the new development with the Autodesk software meant the process was faster than having to repeatedly redraw the designs to focus on a particular factor, such as carbon emissions or cost, said Ryan McNulty, a principal architect at MBH Architects, which also worked on the project.
A human architect could optimize for specific attributes, such as heating or privacy, Benjamin said, while the system “can explore hundreds of options really quickly.”
McNulty said the tool shaved off hours in design time that would otherwise have to be done by hand, adding, “for us it’s a tool to understand what we’re trying to do and make decisions faster and more effectively.” But, he cautioned, it’s “not a silver bullet that allows us to press a button and build a building.”
Autodesk isn’t the only software maker or researcher applying AI to design and construction. Companies including Riveia, HD Lab, and TestFit are also testing various forms of so-called generative design that let creators imagine buildings inside and out.
A project at nonprofit tech research institute SRI International devised a generative AI program to turn sketches into potential architectural designs and more efficiently map textures onto existing buildings, in partnership with Japanese construction giant Obayashi Corp.
SRI Senior Computer Scientist Eric Yeh said his team is working on ways to alter aspects of an AI-generated building design using text prompts. He hopes to come up with tools that can map out an initial building design, including its cost and other attributes, in the near future.
Another company called Augmenta, which was founded by Autodesk alum Francesco Iorio, aims to take 3D models of buildings and use AI to automatically determine the best placement for internal systems such as electrical wiring, factoring in cost, electrical codes, and the availability of parts.
Iorio, who worked on Autodesk’s AI technology originally intended for designing aircraft components — the predecessor to the AI building design software — said automating parts of the construction design stage could also represent huge savings on large, non-residential projects such as hospitals or offices.
Software company Slate in Pleasanton gives construction managers a single database that gathers all the deadlines and moving parts of a project, from subcontractors to open requests for information. It even accounts for weather conditions. CEO Trevor Schick said he hopes to build a software tool by the end of this year that can generate designs based on a set of parameters, along with a detailed project and construction schedule.
From that perspective AI software is seemingly transformative. Yet for housing developers facing slow-moving financing and permitting processes, it may represent only an incremental improvement.
Case in point: The final development permit to begin building on the Phoenix site was approved by Oakland’s planning commission nearly five years ago, according to an email from Jean Walsh, Oakland’s public information officer. The planning approvals were extended twice to give the developers more time, she said.
One longtime nonprofit housing expert was skeptical about how much impact AI could have on affordable housing development.
“The nonprofit housing industry is already unequivocally the most skilled manager of building development,” said Sam Moss, executive director of San Francisco’s Mission Housing Development Corporation, speaking in general and not about the Phoenix project.
Moss conceded that there are opportunities to make the process more efficient. “But in terms of making a giant impact where we’re going to deliver all these magical units a year earlier, it’s way more important to focus on drastically increasing our funding” and pass pro-housing legislation, statewide he said.
At the Phoenix project, construction is moving along. But the project has yet to receive a final development permit for the remainder of its more than 200 market-rate units, according to Walsh, Oakland’s public information officer.
Said Hiteshew, the developer. “We don’t have firm pricing for the balance of phase two, or the total project.”
Reach Chase DiFeliciantonio: chase.difeliciantonio@sfchronicle.com; Twitter: @ChaseDiFelice
Read the full article in the San Francisco Chronicle here.