Leveraging artificial intelligence (A.I.) during and after commercial building construction could markedly reduce energy consumption and carbon emissions, and when combined with energy policy measures and the use of low-carbon power generation, the reductions could be substantial, according to a study published in July in Nature Communications.

Researchers at the Lawrence Berkeley National Laboratory analyzed the energy savings within a typical midsized office building under six situations: a “frozen” scenario using the current building efficiency level as the study’s baseline; two “business-as-usual” scenarios—continued technology improvement and a higher stock of highly efficient and net-zero buildings—with or without A.I.; and three policy-driven situations promoting highly energy-efficient buildings and net-zero energy buildings, and even more aggressive policy implementation to achieve zero emissions by 2050.

The major policy measures within the study include the promotion of efficiency technologies, implementation of building codes and energy-efficiency standards, incentives, subsidies, financial assistance and government-funded programs. The policies were based on the measures outlined in the Biden Administration’s Long-Term Strategy of the United States: Pathways to Net-Zero Greenhouse Gas Emissions by 2050.

Potential energy savings metrics were categorized under equipment, occupancy influence, control and operation, and design and construction. The researchers found that adopting A.I. could reduce energy consumption and carbon emissions by roughly 8% to 19% in 2050.

However, combining A.I. with energy policy and low-carbon power generation, energy consumption could be reduced by as much as 40% and carbon emissions by as much as 90%, compared to business-as-usual scenarios in 2050.

Within the construction of buildings, A.I. can be leveraged to reduce costs, mitigate risks and enhance the health and safety of workers, according to the researchers.

Once occupied, A.I. can produce more savings with smart controls, analyzing system diagnostics, occupancy behavior, load prediction and demand response.

“With advanced A.I. algorithms such as deep learning and reinforcement learning, the A.I. model can itself learn from operational data and evolve itself with continuous live data to optimize objective functions and improve performance,” according to the report.