Scientists have outlined how generative Artificial Intelligence (AI), physics-based modelling and automated experimentation could work together to accelerate the discovery and development of electrochemical energy technologies, including batteries, fuel cells and electrolysers.
Publishing their review in Science Bulletin , the international research team argues that today’s AI tools are already helping researchers predict battery lifetime, screen catalysts and analyse complex electrode structures, but most still act mainly as “predictive assistants”. The next step, they suggest, is to build closed-loop systems that can generate new ideas, test them, learn from feedback and improve continuously.
The review, titled “ Closing the loop with generative AI and automated experimentation in electrochemical energy innovation ,” brings together recent progress in generative AI for molecular and crystal discovery, electrode microstructure design, system-level optimization and large language model applications in electrochemical energy research.
Using examples from batteries, fuel cells and electrolysers, the researchers identify three major ways in which generative AI could reshape energy innovation:
Co-author Dr Zhiqiang Niu, from the University of Birmingham, said: “Electrochemical energy technologies are central to the net-zero transition, but their development is still slowed by enormous design spaces across materials, electrodes and devices. Generative AI offers a new opportunity to move beyond prediction and begin creating new candidate materials, structures and research hypotheses.”
“Our review shows that the real power of generative AI will come when it is connected to physics-based models and automated experiments. This would allow AI-generated ideas to be tested, validated and improved in a continuous closed loop.”
The team proposes a new research framework called Generative Electrochemical Intelligence , or GenE. GenE is envisioned as a physics-informed, multimodal and human-in-the-loop platform that connects generative AI with automated robotic experimentation.
In the proposed GenE framework, four types of AI agents work together:
Together, these agents could form a closed research cycle: AI proposes new materials or structures, modelling tools evaluate their likely performance, robotic platforms test them in the laboratory, and the results are returned to the system for the next round of improvement.
The researchers highlight that such a framework could be particularly valuable for electrochemical energy systems because their performance depends on multiple length scales, from atomic-scale materials and nanoscale electrode structures to full devices and operating conditions.
Dr Niu added: “The aim is not to replace scientists, but to give them more powerful tools. Human researchers will still be essential for setting goals, defining safety limits and making high-impact decisions. GenE is about creating a smarter partnership between scientists, AI models and robotic laboratories.”
The review also warns that major challenges remain before such systems can be widely deployed. Current large language models can produce hallucinations or unreliable outputs, generative models may create physically unrealistic structures, and large AI systems can consume significant computational energy. Automated platforms also raise important questions around safety, responsibility, data ownership and human oversight.
To address these issues, the authors call for stronger physics constraints in AI models, higher-quality cross-domain datasets, uncertainty quantification, more sustainable AI architectures and carefully designed human-in-the-loop safeguards.
In the future, the researchers suggest that closed-loop platforms such as GenE could help accelerate the development of deployable energy technologies for grid storage, electric transport, green hydrogen production and low-carbon power systems. By combining digital creativity with physical validation, generative AI and automated experimentation could shorten the path from scientific idea to real-world energy device.
Science Bulletin