Meta has used AI to develop concrete that emits 40 per cent less carbon, but researchers say that other concrete mixtures can already achieve that
Facebook’s parent company, Meta, has used AI to develop a new way of creating concrete which it claims produces 40 per cent less carbon emissions than standard mixtures, and is already using it in its latest data centre. But experts say that concrete mixtures with similar emissions are already in use across Europe, and that constructing new buildings is incompatible with reducing carbon pollution.
Meta is investing heavily in AI research, including building the world’s most powerful AI-specific supercomputer. Its main aims are to develop better speech-recognition tools, automatically translate between different languages and help build a 3D virtual metaverse, but the company is also using AI to work on projects such as concrete production.
The company says that this construction material is a major contributor to its carbon footprint as it builds data centres around the world for its online services. The production and use of concrete is responsible for around 8 per cent of global carbon emissions.
Basic concrete is a mix of cement and an aggregate such as gravel, mixed with water. Commercial concrete can contain dozens or even hundreds of ingredients to achieve a desired strength or durability. Cement is responsible for a large part of concrete’s carbon emissions, so alternatives to this component with a smaller but still sizeable carbon footprint such as fly ash, a by-product of burning coal, and slag, which is a by-product of manufacturing steel, are commonly used.
To train an AI model to find the optimal recipe for a given use, Meta researchers used a resource which lists various properties of 1030 different concrete mixtures called the Concrete Compressive Strength Data Set. This includes the strength of each concrete after curing for a week and a month, and its carbon footprint. The AI was then able to compare all possible concrete mixtures and find examples that matched a minimum given strength, but also had the lowest possible emissions .
Meta took the five most promising mixtures and tested them in a laboratory. The company said in a blog post that manually discovering such mixtures would be challenging, but that its AI approach completed the task “within weeks”. In a paper outlining the research, the authors say coming up with such mixtures is “pushing the boundary of human creativity”.
Walter Kaufmann at the Swiss Federal Institute of Technology in Zürich says that the claimed reduction in carbon emissions is calculated in comparison with a concrete made using pure Portland cement produced entirely with fossil fuel energy and no replacement material at all. “I doubt that this is US industry standard today – it would be embarrassing if it were,” he says.
The AI-derived mixtures are certainly effective, but not really novel, says Kaufmann. “In western Europe, concretes similar to the one referenced as ‘industry standard’ in the paper stopped being used 25 years ago. The emissions of the ‘novel’ concretes described in the paper essentially correspond to Swiss average concrete in terms of emissions today,” he says.
One of the new concrete mixtures is being used in limited parts of Meta’s new data centre in DeKalb, Illinois. “Based on a life cycle assessment of our typical data centre design, we have concluded that concrete is a significant contributor to our embodied carbon emissions, and hence the need to find low-carbon solutions for concrete,” says Julius Kusuma, a researcher at Meta.
Phil Purnell at the University of Leeds, UK, says that minimising cement content in concrete has been a goal of manufacturers for some time because it is the most expensive part of the mixture, and that minimising carbon emissions in reality means not creating new buildings.
“You can fiddle around at the edges of cement contents, using a bit less energy and all the rest of it, but at the end of the day, if you really want to decarbonise the built environment, stop building stuff. It’s literally that simple,” he says.
More on these topics: