AI has entered the building, but does that make it greener?
Algorithms are finding surprising new ways to make buildings smarter,reduce emissions. But what happens when we factor in the carbon footprint of the AI itself?
How does your building breathe?
Given how much power it takes to keep a large space ventilated, cooled or heated, doing it right can completely alter a structure’s carbon footprint. But what does “right” look like? New AI-led start-ups are offering to help “think” this through in new ways.
A key pilot project that has shown interesting outcomes was conducted at Loyola University’s Schreiber Center in Chicago. This 10-storey mixed-use facility attached to its business school was already a very green building when it was inaugurated, in 2015.
Its gold rating from LEED (Leadership in Energy and Environmental Design) essentially certified that it used about 30% less energy than a traditional building its size would. Much of these savings came from the fact that it was built to capitalise on natural ventilation and natural light.
In 2022, the administration decided to see if they could raise its energy efficiency and reduce emissions further. They reached out to Canadian start-up BrainBox, which took over the Schreiber Center’s HVAC (heating, ventilation and air-conditioning) system with its proprietary, AI-led algorithm.
How could the building possibly cut energy use further? The program found interesting ways to answer that question.
One of these involved “watching” for a surge of renewable energy on the grid. When such a surge occurred, the algorithm raised the HVAC system’s energy use, to warm or cool the building to a greater degree. When the grid switched back to coal- or gas-powered energy, the algorithm could then disconnect the cooling and heating systems for a while, allowing the building to thermally drift.
In addition, the algorithm monitored occupancy and adjusted cooling, heating, lighting and other energy use to fit (as many smart-home systems can do, including, for instance, Google Nest).
In May this year, the results of a year-long assessment were released, and it turned out that Schreiber Center’s energy bills had fallen by 10% and estimated carbon-dioxide emissions, by 15%. (The assessment was conducted in association with University of California Berkeley’s Center for the Built Environment and the clean-energy NGO WattTime.)
How big a difference does the element of AI make? Well, something like Nest, which is led by machine-learning algorithms, can only be reactive, recognising symptoms and picking from a set number of possible responses.
Artificial intelligence can be proactive. “Tell” an AI algorithm to focus on cutting emissions, and it will gather data accordingly, analyse what kinds of options to deploy, and re-order a system if needed.
Some are hoping that AI can eventually be deployed to do this in the construction industry from the earliest stages of design.
Instead of a steel frame, for example, a program might suggest an alternative material that can be procured locally and transported more easily, or that might simply last longer based on local usage patterns and climate, says Des Fagan, head of architecture at Lancaster University and a member of Royal Institute of British Architects’ (RIBA) expert advisory group on AI.
“We’re not quite there yet,” Fagan adds.
Costs of construction could fall too, if BrainBox founder Sam Ramadori is right in suggesting that such programs could all be available for as little as the price of a software subscription.
But, and it’s a big but, there is the footprint of the AI itself. It is still not clear how this would tie into the idea of green buildings and energy savings.
Running an AI program requires so many data centres — involving so much power and water — just at the training stage, that a study by researchers at Carnegie Mellon University and the data-science platform Hugging Face found that generating a single image essentially sucks up as much energy as fully charging a smartphone.
How long would it take for the energy saved in a green building to compensate for running the AI program that made and kept it greener?
It’s still early days for a clear answer, Fagan says. “It’s certainly something we are all conscious of,” he adds. “Moving forward, such calculations will need to part of the formula itself.”