Hao Li, Distinguished Professor at Tohoku University's Advanced Institute for Materials Research, poses a question that challenges how we think about scientific progress: What if the breakthrough your field needs has already been discovered, but remains trapped in an old graph or forgotten table? This is the insight driving a sweeping review published in Chemical Communications, where researchers at the WPI-AIMR demonstrate that the next generation of discoveries may not come from laboratories creating fresh data, but from AI systems learning to see old data in radically new ways.
Every year, science produces more information than researchers can possibly absorb. Thousands of chemistry and materials studies pile up in journals, conference proceedings, and institutional databases—each containing valuable experiments, measurements, and observations that slip from the scientific conversation. The challenge is no longer finding information; it's extracting meaning from an ocean of existing knowledge. Tohoku's team argues this is where artificial intelligence becomes transformative. By combining AI and data science with decades of published research, they show how hidden patterns and connections can emerge, accelerating everything from battery technology to hydrogen storage systems.
The researchers walk through three concrete applications. In catalysis research, data-driven approaches are revealing phenomena that traditional theoretical models missed, allowing chemists to screen and design materials far more efficiently than through conventional trial-and-error. For solid-state electrolytes—materials critical to next-generation batteries—AI-based methods are uncovering the underlying physical mechanisms that govern how ions move through solids. This deeper understanding directly supports the discovery of improved electrolyte materials. The hydrogen storage field offers perhaps the most complete example: researchers traced a pathway from extracting knowledge hidden in old data, structuring it into organized frameworks, and eventually feeding it into autonomous design systems that generate novel storage solutions.
Li emphasizes that this shift represents a fundamental change in how materials research operates. "Scientific discovery is no longer driven only by creating new data," he explains. "Instead of relying on slow trial-and-error methods, the next breakthrough may come from seeing old knowledge in a completely new way with the help of AI." The vision emerging from this work is of a digital materials ecosystem where knowledge extracted from historical studies connects seamlessly with theoretical simulations and new experimental validation, creating feedback loops that compress the timeline for discovering better materials.
What makes this particularly significant is its implications for fields starved of resources. Not every laboratory can afford state-of-the-art equipment or conduct thousands of experiments. But every researcher can access published literature. By developing AI tools that intelligently mine this shared scientific heritage, institutions worldwide gain a more equal footing in the race for breakthroughs. The review suggests that database construction and intelligent agents will become cornerstones of twenty-first-century materials research.
The message is hopeful and humbling in equal measure: the knowledge driving future progress may already exist. What scientists need now is the skill to listen to it differently. As the researchers conclude, in science, everything old can become new again.