Concrete has stood as the cornerstone of construction for millennia, tracing back to ancient civilizations and remaining today as the world's most widely utilized composite material. Despite its enduring popularity, this essential building material carries significant environmental drawbacks that demand innovative solutions.The production of cement, concrete's primary ingredient, generates 8-9 percent of global anthropogenic CO2 emissions while consuming 2-3 percent of worldwide energy—figures projected to rise in coming years. With aging infrastructure across the United States, recent federal legislation has allocated unprecedented funding for revitalization projects, alongside mandates to reduce greenhouse gas emissions wherever possible, positioning concrete as a critical focus for modernization efforts.
Elsa Olivetti, the Esther and Harold E. Edgerton Associate Professor in MIT's Department of Materials Science and Engineering, alongside Jie Chen, MIT-IBM Watson AI Lab research scientist and manager, believes artificial intelligence holds the key to addressing these challenges. Their pioneering work leverages AI to design and formulate next-generation concrete mixtures that significantly reduce costs and carbon emissions while enhancing material performance and incorporating manufacturing byproducts. Olivetti's expertise in materials sustainability combined with Chen's advanced machine learning techniques has created a powerful synergy. Together with their collaborators, they've launched an ambitious MIT-IBM Watson AI Lab project aimed at revolutionizing concrete sustainability for the benefit of society, climate, and economy.
Q: What makes concrete such a prevalent building material, and where is it typically used?
Olivetti: Concrete dominates the global construction landscape with an astonishing annual consumption of 30 billion metric tons—more than 20 times the production volume of steel, the second-most produced material. This massive scale inevitably creates substantial environmental impact, contributing approximately 5-8 percent of worldwide greenhouse gas (GHG) emissions. Its popularity stems from several key advantages: it can be manufactured locally using readily available materials, offers versatility across countless structural applications, and remains remarkably cost-effective. Essentially, concrete combines fine and coarse aggregates, water, cement binder (acting as the adhesive component), and various specialized additives to create its final form.
Q: What are concrete's main sustainability challenges, and what specific problems is your project addressing?
Olivetti: The research community is exploring multiple strategies to reduce concrete's environmental footprint, including implementing alternative heating fuels for cement production, enhancing energy and material efficiency at manufacturing facilities, and developing carbon capture technologies. However, one particularly promising avenue focuses on creating alternatives to traditional cement binders. Although cement comprises only 10 percent of concrete's mass, it's responsible for approximately 80 percent of its greenhouse gas emissions. This impact stems from two primary sources: the fuel burned to achieve the high temperatures required for the chemical reaction, and the CO2 released during limestone calcination—the core chemical process itself. Partially replacing conventional cement ingredients (typically ordinary Portland cement or OPC) with alternative materials derived from industrial waste and byproducts can significantly reduce GHG emissions. However, simply substituting these materials doesn't automatically guarantee sustainability. Waste materials might require extensive transportation—adding fuel emissions and costs—or need energy-intensive pretreatment processes. The optimal approach varies considerably depending on local conditions and available resources. Furthermore, given the enormous scale of concrete production globally, any viable solution must accommodate the massive volumes required. Our project aims to develop novel concrete formulations that substantially reduce cement and concrete's GHG impact, moving beyond traditional trial-and-error methods toward more predictive, data-driven approaches.
Chen: As we confront the urgent challenge of climate change, we must ask: can we identify alternative ingredients or reformulations that significantly reduce greenhouse gas emissions? Through this machine learning initiative, we're working to discover scientifically grounded answers that balance environmental responsibility with practical application.
Q: Why is addressing concrete's environmental impact particularly urgent now?
Olivetti: The climate crisis demands aggressive, immediate action to reduce greenhouse gas emissions across all sectors. While transportation and electricity generation have relatively clear pathways to decarbonization, industrial materials production presents more complex challenges. These "hard-to-abate" sectors lack well-defined roadmaps for emissions reduction, making them particularly critical targets for innovation. As other industries implement cleaner technologies, the relative impact of concrete production becomes increasingly significant, creating an urgent need for sustainable alternatives that can be implemented at scale.
Q: How does your approach specifically aim to create more sustainable concrete?
Olivetti: Our primary objective is to develop predictive models that identify concrete mixtures capable of meeting critical performance requirements—such as strength and durability—while simultaneously optimizing economic and environmental factors. A central strategy involves incorporating industrial waste materials into blended cements and concrete formulations. Success requires deep understanding of the glass and mineral reactivity of these constituent materials. This reactivity not only determines the maximum viable replacement levels in cement systems but also influences concrete processing characteristics, strength development, and pore structure formation—all of which ultimately control concrete durability and life-cycle CO2 emissions.
Chen: We're systematically investigating how various waste materials can partially replace cement components—a strategy we hypothesize will enhance both sustainability and economic viability. Waste materials are typically abundant and inexpensive, making them attractive alternatives. By reducing cement content, the resulting concrete product generates significantly less carbon dioxide. However, determining the optimal mixture proportions that produce durable concrete while achieving multiple objectives presents an enormously complex challenge. Machine learning offers unprecedented opportunities to advance predictive modeling, uncertainty quantification, and optimization techniques to address this multifaceted problem. We're exploring deep learning approaches combined with multi-objective optimization algorithms to identify promising solutions. These computational methods have recently become sufficiently sophisticated to deliver reliable results with quantified uncertainty—essential characteristics for developing truly sustainable concrete formulations.
Q: What specific artificial intelligence and computational techniques are you employing in this research?
Olivetti: We're implementing advanced AI techniques to gather comprehensive data on concrete ingredients, mixture proportions, and performance characteristics from scientific literature using natural language processing. We supplement this information with data from industry partners and high-throughput atomistic modeling and experiments to optimize concrete mixture designs. This combined dataset enables us to develop insights into the reactivity of potential waste and byproduct materials as cement alternatives for low-CO2 concrete formulations. By incorporating fundamental information about concrete ingredients, our resulting performance predictors should prove more reliable and transformative than existing AI models in the field.
Chen: Our ultimate objective is to determine the optimal constituents and their precise proportions for concrete formulations that balance multiple factors: strength, cost, environmental impact, and performance characteristics. Each of these objectives requires specialized modeling approaches: we need models to predict concrete performance (including longevity and load-bearing capacity), models to estimate production costs, and models to calculate carbon dioxide generation. We're constructing these models using diverse data sources from scientific literature, industry partners, and laboratory experiments. We're exploring Gaussian process models to predict concrete strength development over days and weeks. These models provide valuable uncertainty estimates alongside their predictions, which is crucial for decision-making. However, these models require specific parameter specifications, which we calculate using additional modeling approaches. Simultaneously, we're investigating neural network models because they allow us to incorporate domain knowledge from human experts. Our model architecture ranges from relatively simple multi-layer perceptrons to more complex graph neural networks. The goal is developing models that are not only accurate but also robust—capable of handling noisy input data while maintaining reliable predictions for multi-objective optimization purposes. Once we've developed sufficiently reliable models, we'll integrate their predictions and uncertainty estimates into multi-objective optimization frameworks that account for various constraints and uncertainties.
Q: How do you navigate the complex trade-offs between competing objectives?
Chen: The multiple objectives we consider aren't necessarily aligned—sometimes they directly conflict with each other. Our goal is identifying Pareto-optimal scenarios where one objective cannot be improved without compromising others. For instance, further reducing costs might necessitate accepting lower performance or increased environmental impact. Ultimately, we'll present these findings to policymakers who must weigh various factors based on their specific priorities and constraints. For example, they might accept slightly higher costs in exchange for substantial greenhouse gas reductions. Conversely, if minimal cost variations produce dramatic performance improvements—doubling or tripling certain metrics—this would represent a clearly favorable outcome worth pursuing.
Q: What significant challenges have you encountered in this research?
Chen: The data we obtain from industry sources or scientific literature presents considerable challenges. Concrete measurements can vary significantly depending on collection timing and methodology. Additionally, integrating data from diverse sources creates substantial missing data problems that require extensive preprocessing to make suitable for machine learning model development and training. We're exploring various imputation techniques to substitute missing features, alongside developing models that can tolerate incomplete information in our predictive modeling and uncertainty estimation frameworks.
Q: What outcomes do you hope to achieve through this research?
Chen: Ultimately, we aim to provide manufacturers and policymakers with either specific concrete recipes or a continuum of formulation options that balance multiple objectives. We believe this research will deliver invaluable insights for both the construction industry and global environmental protection efforts.
Olivetti: We're working to develop robust methodologies for designing cements that effectively incorporate waste materials to reduce their CO2 footprint. Since waste generation isn't standardized, we cannot depend on a single consistent feedstock if we want our solutions to achieve massive scalability. Our approaches must remain flexible and adaptable to accommodate varying feedstock characteristics—requiring deeper fundamental understanding. Our strategy involves developing local, dynamic, and flexible alternatives by identifying the fundamental properties that make waste materials reactive, enabling us to optimize their utilization across diverse scenarios. We accomplish this through predictive model development using specialized software my team has created to automatically extract data from over 5 million scientific texts and patents across various domains. We connect this data infrastructure with the innovative capabilities of our IBM collaborators to design methods that predict the final impact of novel cement formulations. If successful, we can significantly reduce emissions from this ubiquitous material while contributing meaningfully to global carbon mitigation goals.
Additional researchers contributing to this project include Stefanie Jegelka, the X-Window Consortium Career Development Associate Professor in MIT's Department of Electrical Engineering and Computer Science; Richard Goodwin, IBM principal researcher; Soumya Ghosh, MIT-IBM Watson AI Lab research staff member; and Kristen Severson, former research staff member. Collaborators also included Nghia Hoang, former research staff member with MIT-IBM Watson AI Lab and IBM Research, and Executive Director of MIT Climate & Sustainability Consortium Jeremy Gregory.
This research receives support from the MIT-IBM Watson AI Lab.
