Over the past two decades, computational approaches have taken on a complementary role alongside traditional experimental synthesis and characterization workflows in materials science. However, experimental approaches are costly and time-consuming, while computational methods require extensive computational power, limiting efficient material discovery and analysis. Additionally, predicting material properties often relies on intuition and experience.

Artificial intelligence (AI) has emerged as a promising approach to address these limitations. By uncovering patterns in large datasets, AI-based methods enable more efficient exploration and interpretation of material behavior across both computational and experimental domains, with the potential to significantly shorten discovery timelines.

To explore this emerging field of materials informatics, a recent review published in the Journal of the Physical Society of Japan surveys current progress and outlines key challenges in integrating AI into materials science.

The review explores several AI-based applications in materials science, including acceleration of first-principles calculations and computational models, utilization of generative models for material design, and the use of large language models both as guides and knowledge extractors. It also discusses approaches for optimizing experimental conditions and for integrating computational and experimental data.

A central contribution of the study is the identification of a fundamental epistemological divergence that hinders effective AI integration. In traditional materials science, a single breakthrough discovery may be considered highly valuable, whereas machine learning and data modeling approaches require systematically varied datasets across many materials. In addition, these paradigms rely on different reasoning methodologies.

This rift has created data biases, quality inconsistencies, interpretability problems, and communication barriers. To resolve this epistemological barrier, the study proposes concrete, implementable solutions, including data-bias correction strategies, physics-informed model architectures, and explainable AI frameworks. It also emphasizes the importance of data standardization and interdisciplinary knowledge-transfer mechanisms.

Overall, the review provides a comprehensive perspective on AI integration in materials science and outlines directions that may support continued progress in the field.

Share this topic

Fields