AI in Buildings: Where the Biggest Impact Lies, and What Mud Can Teach Us

Asked about AI's effect on architecture, people picture towers of strange shapes. The greater effect, we believe, is far quieter: existing buildings that consume less, planning decisions tested before they are carried out, and a climatic heritage that is not squandered. This view has its evidence and its limits.

The scale of the matter

The report by the UN Environment Programme and the Global Alliance for Buildings and Construction (March 2025) stated that the sector contributes about 34 per cent of global carbon-dioxide emissions and consumes around 32 per cent of energy, and that nearly half of the buildings that will exist in 2050 have yet to be built. This means opportunities for improvement are divided between what we build now and what has already been built, and improving an existing building is no less important than designing a new one. The prior question remains: what do we ask of the tool? Energy efficiency alone, or efficiency that preserves fairness of access for people and identity for the place?

What studies say about generative design

In generative design the architect sets goals and constraints, and the computer proposes and ranks many alternatives. A systematic review of 34 studies on its application to energy efficiency in buildings found that the tools reduced consumption compared with prototypes or traditional design: by up to 28 per cent in thermal performance and 23.3 per cent in energy performance. The review observes that the knowledge required is complex, and that evolutionary algorithms are the most widely used.

These figures should be read cautiously: they are the results of cases mostly studied through simulation, and a building's performance once people live in it may differ from its performance on paper. The International Energy Agency likewise notes that the potential in buildings is limited by the pace of digitalisation. So do not take a single figure as the yardstick for every building.

What mud teaches us

In 2010 UNESCO inscribed the At-Turaif District in ad-Dir'iyah on the World Heritage List, describing the Najdi style as specific to the centre of the Arabian Peninsula, built with mud (adobe) in the great palaces and well adapted to its environment. A recent study (2022) simulated two traditional Najdi houses, in Riyadh and Hail, calibrated against field measurements, and added a number: replacing mud-brick walls with modern ones raised consumption by about 33 per cent in Riyadh and about 9.5 per cent in Hail, and raising the window-to-wall ratio raised it by about 9 per cent in Riyadh. These are simulation results, not post-occupancy measurements, and limited to two cases, but they suggest that old climatic wisdom can be measured.

The practical benefit is to bring this heritage into generative tools as criteria and data. And this is our own theoretical warning, not a measured result: models that learn from an archive of existing buildings may lean towards the common, and so weaken the distinctiveness of a place unless we feed them our local knowledge.

What to do

• Direct the tool at existing buildings: predictive maintenance, and tuning air-conditioning and lighting to actual use.
• Compare paper with reality: measure the building's performance after occupancy and feed it back into the model.
• Make fairness and identity written criteria: not efficiency alone, otherwise efficient cities without soul emerge.
• Teach the architect to engage the tool critically: not to accept its output unexamined.
• Place city data under transparent governance, so that it does not turn into a surveillance tool (see the article "Privacy in the Age of AI: How Your Secrets Are Inferred, and How to Protect Them").

It is a fitting reminder that the question of what makes good shelter has not yet been settled; all the algorithm has done is put it to us again with urgency: which city do we want, and which city do we leave to those after us? The answer is in our own hands; we take the means and rely on God.