A team at the University of Rochester has figured out how to use artificial intelligence to dramatically speed up the discovery of materials that could turn carbon dioxide into fuel, potentially transforming how we approach carbon capture and clean energy. Instead of running hundreds of thousands of experiments, researchers used large language models (LLMs) like ChatGPT to generate step-by-step recipes for creating new catalysts, reducing the number of experiments needed from 360,000 to just 10 .

How Can AI Language Models Speed Up Materials Discovery?

The breakthrough comes from a counterintuitive insight: instead of asking AI to predict complex numerical data about materials, researchers asked it to describe how to make them. Marc Porosoff, an associate professor in the Department of Chemical and Sustainability Engineering at the University of Rochester, likened the approach to describing coffee. You could describe it by taste, color, and aroma, or you could describe the recipe: the type of beans, grind size, water temperature, and brewing method. Both describe the same cup of coffee, but the recipe is something others can actually replicate .

"We're able to leverage the pre-trained knowledge of large language models and well-established statistical methods for materials discovery to help us as researchers navigate large experimental design spaces more efficiently," said Marc Porosoff.

Marc Porosoff, Associate Professor, Department of Chemical and Sustainability Engineering, University of Rochester

The method works by having researchers input natural language prompts describing the materials they want to create. The AI then suggests optimal procedures for experiments. As researchers run the experiments and input the results back into the AI model, they continue iterating until they reach their goal. This feedback loop dramatically reduces the search space .

Traditional AI methods for materials discovery rely on something called Bayesian optimization, which identifies the best candidates by exploring a massive parameter space. However, the results are complex numerical data that requires deep expertise to interpret and use. The new LLM method instead produces procedures that any trained researcher can understand, execute, and verify .

Why Does This Matter for Climate and Clean Energy?

The implications for climate action are significant. One of the most promising applications is creating catalysts that convert carbon dioxide and hydrogen into useful chemicals like methanol and ethanol. These can be used as fuels or as ingredients in pharmaceuticals, cosmetics, and other products. Currently, it takes a decade or longer to go from conceptualizing a new catalyst to testing it in a real reactor. The new AI method could compress that timeline to a single year .

The U.S. Department of Energy's Advanced Research Projects Agency-Energy (ARPA-E) recognized the potential and announced nearly $3 million in funding for the University of Rochester team to apply this method toward creating catalysts for fuel production from abundant materials. The project, called CATALCHEM-E (Catalytic Application Testing for Accelerated Learning Chemistries via High-throughput Experimentation and Modeling Efficiently), is scheduled to begin in July and run through 2029, involving a multi-institution team including Virginia Tech, Stanford University, Northwestern University, and others .

How to Understand the Real-World Impact of AI Energy Use

While AI is helping solve climate problems, it's also important to understand the energy footprint of AI itself. At Michigan State University, librarians created an interactive exhibit called "How Green is Your Chat?" to help people visualize the water and energy consumption associated with using AI chatbots like ChatGPT and Gemini .

• Interactive Tool: Visitors could select an AI provider and specify what task they were using it for, from typing a tweet to writing a book report, and see the corresponding water and energy impact displayed on a large screen.
• Local Impact Awareness: The exhibit included a satellite map of Michigan showing proposed data center locations across the state, helping people understand how AI infrastructure affects their own communities.
• Educational Approach: The event used unconventional methods, including real cookies, to create a bridge between individual AI usage awareness and the systemic impacts of building data centers that require massive resource extraction.

"I think we just wanted to have a tool that could show in a very direct way the energy impacts in using generative AI, and one of the main things that we use is a cookie here. We're trying to create a bridge between making people aware of their individual usage, and then what the more systemic impacts of that usage is in terms of building data centers and creating an extractive model for resources to support this technology," said Justin Wadland.

Justin Wadland, Head of Digital Scholarship Services, MSU Libraries

This dual reality reflects the current state of AI and climate: the technology can accelerate solutions to environmental problems, but it also demands significant energy and water resources. The Rochester team's approach to materials discovery is one example of AI being used to solve problems faster and more efficiently, potentially reducing the total experimental energy required to develop climate solutions .

The research demonstrates that the barrier to using advanced AI for scientific discovery isn't just computational power; it's also about making the technology accessible to researchers who may not have deep expertise in machine learning. By translating complex optimization problems into human-readable recipes, the team has democratized access to AI-powered materials discovery .

Shane Michtavy, a chemical engineering PhD student who helped develop the method, noted that using pre-trained LLMs allows researchers to explore using less data than traditional models, since the AI comes with built-in knowledge of the physical world and catalysis . This efficiency gain could have ripple effects across the field, enabling more researchers at more institutions to participate in the search for climate solutions.