With a Little Help from AI

Concrete is the world’s most used building material with more than 10 billion tons produced annually and production at this scale is a significant resource burden. Manufacturing concrete uses excessive energy and water while also producing greenhouse gasses and other pollutants. Overall, the production of concrete contributes to 8% of global carbon emissions. A team of researchers from Meta, IBM and Ozinga with researchers from USC, University of Illinois Urbana-Champaign and University of Chicago set out improve concrete.

They ran an experiment to determine if artificial intelligence could create a less carbon-intensive concrete that maintains structural integrity. As a material, concrete is composed of air, water, sand/gravel, and cement. The ratio of the four constituents and physical variations in them, determine the building characteristics of the final concrete. With the goal of meeting engineering performance requirements and reducing environmental impact, the team designed the experiment. A generative AI model was trained using environmental impact data and a small public dataset. Employing semi-supervised machine learning, the model was able to come up with different formulations that met all of the researchers requirements:

• reduced carbon footprint

• significant compressive strength

• comparable durability

The experiment concluded that AI was able to discover formulations that could cut total carbon emissions by about half. To test the results Ozinga, a concrete supplier, used the new formulations to produce and test the material. This improved and more environmental friendly version of concrete was used in the construction of a guard tower and a building at one of Meta’s data centers, in DeKalb, Illinois.