In an MIT lab abundant with scribbled notes, complex chemical equations, and doodled diagrams, a persistent research team immersed themselves in a pressing challenge. The researchers from Olivetti Group and the MIT Concrete Sustainability Hub (CSHub) sought a viable solution to reducing the cement content in concrete. By doing so, they could drive down both the material’s costs and carbon emissions, a significant step towards a more sustainable construction industry.
The negative environmental impact of cement is no fresh news. The construction sector has been supplementing cement with materials like fly ash and slag, byproducts of coal and steel manufacturing. Despite aiding in the lowering of emissions, the constant rise in demand makes it a challenge to maintain supply. With countless potential alternatives, the real hurdle lies in the sheer volume of options and the lack of time and resources to properly evaluate each of them.
Enter artificial intelligence (AI) – the game-changer. On May 17, the team, led by Soroush Mahjoubi, a postdoctoral researcher, published an enlightening paper in Nature’s Communications Materials that showcased a new approach. This paper introduced how AI, particularly large language models (LLMs), might drastically revamp how we recognize viable cement substitutes. “We realized that AI was the key to moving forward,” revealed Mahjoubi. “AI lets us wade through the massive volume of scientific literature efficiently without missing key data.”
The team’s AI-powered framework sifted through potential materials, assessing them based on two crucial properties: hydraulic reactivity and pozzolanicity. The former certifies that a substance can harden when introduced to water, similar to traditional cement. Pozzolanicity, in contrast, ensures a substance reacts with calcium hydroxide (a leftover from cement hydration) to strengthen the concrete over time. A material’s fitness as a suitable substitute hinges significantly on balancing these two aspects. “It’s not merely about finding replacements — it’s about ensuring the final product performs comparably or even better,” Mahjoubi stressed.
Using their innovative framework powered by AI, the team analyzed over a million rock samples and a colossal data trove of scientific information. The team grouped possible cement alternatives into 19 unique categories; these ranged from agricultural debris, industrial byproducts, and materials salvaged from demolished buildings. The bunch included old bricks, tiles, and even pottery – materials with high reactivity and require minimal processing to be incorporated into the concrete mix. Mahjoubi shared that some of the most promising candidates were ceramics, much like those found in ancient Roman concrete used for enhancing longevity and water resistance.
As we dig deeper into this research, a broader scope becomes apparent. The innovative approach to creating new infrastructure from waste materials doesn’t simply serve as recycling. The team’s approach truly exemplifies an application of circular economy principles to one of the world’s most widely utilized materials, cement. This remarkable research could very well establish a formidable paradigm for how we manage construction waste. Rather than ending up in landfills, old construction debris could be repurposed to create a new, more sustainable variant of concrete.
But the team’s plan doesn’t end here. They aim to continue refining their AI model to evaluate more types of materials and validate the most promising options through rigorous lab testing. “AI technologies have propelled us far,” stated Professor Elsa Olivetti, senior author of the paper and a mission director for the MIT Climate Project. “We’re excited to see how advances in large language models will push this work even further.” Randolph Kirchain, CSHub director and co-author, highlighted the far-reaching vision: “Concrete is the backbone of the built environment. By deploying data science and AI in material design, we can catalyze the construction industry to build more sustainably — without compromising on strength, safety, or durability.”
This sterling endeavor also saw crucial contributions from MIT postdoc Vineeth Venugopal, Ipek Bensu Manav SM ’21, PhD ’24, and CSHub Deputy Director Hessam AzariJafari. If you are keen on diving deeper into the original research, head to the original release from MIT here: https://news.mit.edu/2025/ai-stirs-recipe-for-concrete-0602
.
Diese Website verwendet Cookies.