Yale and Microsoft researchers are working together to improve redox flow batteries, a technology seen as a safe and scalable option for storing renewable energy on the grid. These batteries store energy in liquid chemical solutions in tanks rather than in solid materials. Researchers are seeking molecular compounds that are stable, highly soluble in water, and capable of delivering high voltage, while also being practical to synthesize in the lab.
To speed that search, David Kwabi and his team at Yale used Microsoft’s CLIO, short for cognitive loop via in-situ optimization. Kwabi’s lab focuses on aqueous organic redox flow batteries, which are considered strong candidates for sustainable and long-duration energy storage. The challenge is balancing stability, voltage, solubility, and efficiency while identifying molecules that can realistically be made and tested. CLIO is designed to reason through scientific problems by reflecting on progress, generating hypotheses, and evaluating multiple discovery strategies, with the ability to adjust when early ideas do not hold up.
Kwabi described the collaboration as a division of labor between human experimentalists and Artificial Intelligence. Humans generate high-quality laboratory data, while Artificial Intelligence can map large chemical design spaces and identify regions worth exploring. In practice, the team starts with a molecular scaffold for a flow battery candidate. CLIO proposes promising molecules, researchers test them in the lab, and the resulting data is returned to the system so it can form new hypotheses and refine its recommendations.
In the project, CLIO first suggested a benzocinnoline-family compound substituted with a benzylphosphonate group. That molecule performed well in some respects but was inefficient at releasing stored energy. After the Yale team fed those results back into the system, CLIO proposed a modified version of the molecule. That revised candidate proved successful, suggesting that the iterative process could significantly improve discovery. Kwabi said the work establishes a new framework for battery science by combining experiment-driven human judgment with Artificial Intelligence’s ability to explore vast chemical design spaces and adapt to experimental evidence.
