The
Romance
of
Research Hao
Li

Climate change, high energy costs and demand, and a lack of food and clean water require urgent action, and we need to accelerate the discovery of new, industrially important materials to combat these problems.

What sparked your interest in materials science?

The answer is simple: materials science can change the world. To tackle climate change, we need to discover new materials that help make engineering more economical and less polluting. New catalysts, batteries, and hydrogen storage materials lay the pathway to a greener, brighter, and cleaner future. I want to play my part in building this future.

Much of your research involves finding new catalysts. Why is discovering new catalysts important for society?

Catalysts possess a wide array of functions. Some are used in fuel cell applications, helping convert hydrogen energy into electricity, and some are used to induce the hydrogen evolution and oxygen evolution reactions. These key reactions drive water splitting, which enables hydrogen to be produced from water. Future green energy sources are underpinned by reactions driven by catalysts, hence why their discovery is so important.

Catalysts are also important for reducing carbon dioxide (CO₂). CO₂’s stability makes it difficult to independently transform it into other compounds. Catalysts make chemical reactions easier and faster. Therefore, my research often investigates new catalysts that help break down CO₂ into useful products like hydrocarbons or carbon monoxide. These products can then be reused in industrial processes.

Lastly, catalysts are also important in water treatment. They help remove, reduce, or even eliminate pollutants in water by converting toxic substances or chemicals into less toxic, or even non-toxic, products.

In a lot of your papers, AI is employed to discover new materials. How is AI revolutionizing the field of materials science?

I am incredibly excited about AI. Traditionally, people spent years, even decades, trying to find suitable materials. It was not unheard of for people to spend their entire lives trying to discover whether one class of materials works well or not. Materials science was notorious for its slow-discovery process.

However, the challenges of today do not afford us the luxury of such a slow pace of discovery. Climate change, high energy costs and demand, and a lack of food and clean water require urgent action, and we need to accelerate the discovery of new, industrially important materials to combat these problems. AI is the clearest path to doing so and why I am so passionate about using it.

However, AI is not a silver bullet. AI requires two important factors if it is to be successfully introduced into the research process. First, AI needs high-quality databases. Like the human brain, if AI is not fed proper information, it cannot generate new and accurate knowledge. That is why I concentrate much of my time on building high-quality databases.

The second important aspect for AI is developing transparent systems. AI is sometimes a black box. But AI can still produce incorrect results. Therefore, we need to know not only what AI predicts, but how it predicts. Only then can we tweak its physical framework to better reflect the physical world and generate higher-quality research.

You make an interesting point. Is this why you pioneered the Digital Catalysis Platform?

Actually, the idea to create a database struck me many years ago. I read about a new material exhibiting high-performance in a high-impact journal. Although the authors claimed it was the first discovery of such a material, a lingering feeling told me this was not the case. It transpired I was right; the material had been written about in a paper nearly 30 years prior. From then on, I wanted to develop a database that could perform large-scale data mining of literature. That way, people can browse the vast treasure troves of information produced throughout human history, leading to less bias.

But my quest for innovative databases did not stop there. I learnt that revisiting old findings can lead to new conclusions. A typical paper, for example, usually measures a material under specific conditions. But if we gather 10, or 100, or even 1000 papers that explore the same material under different conditions, we can draw new discoveries. In that sense, my databases create new insights from old findings.

Now, with the development of large language models, we have a new way to utilize our databases. We can train and develop high-performance, highly precise AI agents by integrating large-scale models, computational backends, databases, and physical frameworks. By combining real-world data and physical frameworks with large-scale AI models, we can achieve the level of accuracy needed for precise analysis and prediction in materials science. The advances are so great they are no longer simply databases, but digital interfaces where practical usage becomes easier thanks to AI.