For several weeks, the laboratory whiteboard overflowed with intricate diagrams, chemical equations, and notes. A dedicated research team from the Olivetti Group and the MIT Concrete Sustainability Hub (CSHub) tackled a pressing challenge: how to minimize cement usage in concrete to lower costs and reduce emissions.
This issue is not a recent phenomenon; alternatives like fly ash—a byproduct of coal combustion—and slag, a byproduct of steel production, have historically replaced portions of cement in concrete. Yet, the rising demand for these materials is outstripping available supply as industries aim to lessen their environmental impact. This urgency led the researchers to a surprising realization: the challenge lay not in the lack of potential substitutes, but in the overwhelming number of choices.
On May 17, the team, spearheaded by postdoctoral researcher Soroush Mahjoubi, released an open-access study in Nature’s Communications Materials highlighting their innovative approach. “We figured out that AI was crucial for progress,” mentions Mahjoubi. “The sheer volume of data regarding potential materials—hundreds of thousands of scientific papers—would take lifetimes to sift through. By the time one finished, new materials would have already emerged!”
Utilizing large language models, akin to the chatbots many of us engage with daily, the team developed a machine-learning framework to evaluate and categorize candidate materials based on their physical and chemical characteristics.
“The first factor is hydraulic reactivity. Cement functions as the ‘glue’ in concrete, solidifying when it interacts with water. Thus, any replacements must behave similarly,” explains Mahjoubi. “Second, we consider pozzolanicity, which describes how a material reacts with calcium hydroxide—a byproduct formed when cement mixes with water—enhancing concrete’s strength over time. We aim to strike a balance between hydraulic and pozzolanic materials for optimal performance.”
By analyzing extensive scientific literature and over 1 million rock samples, the team used their framework to classify potential materials into 19 categories, spanning from biomass and mining byproducts to recycled construction debris. Mahjoubi and his colleagues discovered that numerous viable materials exist worldwide—and many could be integrated into concrete mixes simply by grinding, making it possible to achieve cost reductions and lower emissions with minimal additional processing.
“Interestingly, ceramics present compelling options for substituting cement,” shares Mahjoubi. “Old tiles, bricks, and pottery exhibit significant reactivity. Similar practices were observed in ancient Roman concrete, where ceramics were utilized to waterproof structures. Professor Admir Masic, who leads ancient concrete studies at MIT, and I have had some fascinating discussions regarding this.”
The discovery of everyday and industrial materials like ceramics and mining tailings illustrates how concrete can facilitate a circular economy. By recognizing and reusing materials that would typically contribute to landfill waste, researchers and industry leaders can repurpose them into our infrastructure and buildings.
Looking forward, the research team intends to enhance the framework to accommodate an even broader range of materials, while experimentally validating some of the top candidates. “AI tools have rapidly advanced this research, and we are eager to see how recent advancements in large language models pave the way for future developments,” expresses Professor Elsa Olivetti, the senior author and member of the MIT Department of Materials Science and Engineering. She also holds the role of mission director for the MIT Climate Project and serves as a principal investigator at CSHub.
“Concrete forms the backbone of our built environment,” remarks Randolph Kirchain, co-author and CSHub director. “By leveraging data science and AI tools in material design, we aspire to aid industry efforts in constructing sustainable solutions without sacrificing strength, safety, or durability.”
In addition to Mahjoubi, Olivetti, and Kirchain, other contributors to the study include MIT postdoc Vineeth Venugopal, Ipek Bensu Manav SM ’21, PhD ’24, and CSHub Deputy Director Hessam AzariJafari.
Photo credit & article inspired by: Massachusetts Institute of Technology
