KAIST-led study outlines AI's role in discovery, development, and optimization

An era has now begun where artificial intelligence (AI) can imagine and predict the structures of new materials like humans. Beyond being a mere computational tool, AI now serves as a researcher's 'second brain,' participating in everything from idea generation to experimental validation.

The research team led by Professor Hong Seung-beom of the Department of Materials Science and Engineering at Korea Advanced Institute of Science and Technology (KAIST), in collaboration with U.S. institutions including Drexel University, Northwestern University, the University of Chicago, and the University of Tennessee, announced on the 26th that they had comprehensively analyzed AI's evolution. The research results were published in the international academic journal *ACS Nano* in July.

The team divided the material research process into three stages -- discovery, development, and optimization -- and specifically outlined AI's role in each. AI first recommends promising candidates from vast material databases, reduces trial-and-error in experiments, and autonomously adjusts conditions to find optimal outcomes.

The researchers analyzed how cutting-edge technologies such as generative AI, graph neural networks, and transformer models are transforming AI from a simple computational tool into a thinking researcher. AI learns principles of physics and chemistry on its own to imagine new materials and collaborates throughout the entire process, from proposing ideas to experimental validation.

The study also introduced cases of 'autonomous laboratories,' where AI designs experimental plans and robots perform experiments, as well as 'AI-based catalyst exploration platforms.' In these systems, AI designs experimental conditions, robots automatically synthesize catalysts, and results are analyzed. This technology, which accelerates research, can be applied to other fields such as batteries and energy materials.

However, the team noted that AI's predictions are not always correct. Challenges remain, including data quality imbalances, difficulties in interpreting results, and issues integrating disparate datasets. Therefore, they emphasized that future advancements must focus on AI understanding physical principles and enabling researchers to transparently verify its processes.

Professor Hong Seung-beom stated, "This study demonstrates that AI is becoming not just a tool but a new language and way of thinking in materials science." He added, "The roadmap presented here will provide crucial direction for researchers in fields such as batteries, semiconductors, and energy materials."

Reference

ACS Nano (2025), DOI: https://doi.org/10.1021/acsnano.5c04200