Artificial intelligence (AI) will accelerate materials discovery, but human expertise and education will remain central to responsible and sustainable innovation.
New materials serve as the foundation for major technological advances, providing important advances in areas such as next-generation electronics, robotics, and medical devices. Traditionally, their development has relied on costly, time-consuming, and labor-intensive trial-and-error research. Additionally, the pace of new materials discovery is hampered by the vast design space. As a result, the average timeline for bringing new materials from initial concept to commercialization is typically 10 to 20 years. ¹
How AI is accelerating the pace of research and discovery
AI methods can now predict, discover, and optimize materials much faster and more efficiently. For example, Google’s Graph Networks for Materials Exploration (GNoME) deep learning tool predicted 2.2 million new crystals and identified about 380,000 as stable materials. ² Already, 736 of them have been synthesized by researchers, validating the predictive power of AI. Additionally, the AI-powered autonomous synthesis system was able to generate 41 new compounds in just 17 days.
Structural Constraint Integration in GEnerative Models (SCIGEN) generated over 10 million candidate materials with specific lattice structures associated with quantum properties, of which 1 million passed stability screening. ⁴ Two novel compounds TiPd0.22Bi0.88 and Ti0.5Pd1.5Sb were synthesized and confirmed to exhibit paramagnetic and diamagnetic behavior, demonstrating the ability of AI to bridge computational and experimental design reality. However, accelerating discovery is only the first step to innovation. Translating AI predictions into manufacturable materials remains dependent on processing, manufacturing, and economic analysis that requires expert judgment and cross-disciplinary coordination.
AI accelerates discoveries, but these advances must be rooted in the atomistic understanding provided by physically-based materials simulations. Traditional density functional theory (DFT) and molecular dynamics (MD) simulations are powerful for describing and predicting atomic-level material properties, but are computationally expensive and limited in scale. Machine learning interatomic potentials (MLIP) trained on large datasets such as OMol25 contain over 100 million DFT evaluations across approximately 83 million unique molecular systems and can achieve near-DFT accuracy at significantly reduced computational costs. ⁵ The optimized framework provides faster throughput for MD tasks, further narrowing the gap between high-accuracy simulation and practical ease of use. However, achieving DFT-level accuracy across complex systems remains an active research topic. Together, AI-enhanced DFT and MD will redefine atomistic modeling and accelerate exploration of thousands to millions of candidate structures across energy, catalysis, batteries, and biomaterials domains.
AI can close the loop between simulation and experimentation through fully autonomous, data-driven workflows. The Autonomous Experimentation and Self-Driving Lab (SDL) describes a system where AI, robotics, and automated processes work together in a closed-loop system to accelerate scientific research. For example, recent dynamic flow SDLs have acquired 10 times higher resolution reaction data in record time, making it possible to identify promising inorganic materials in a single pass, significantly reducing the total number of experiments and wasting time and materials. ⁶ Another advance shows self-monitoring robotic systems mapping semiconductor properties. Over a 24-hour period, the system autonomously drove probes across 3,025 prediction points.⁷ These examples demonstrate how SDL can transform discovery timelines, making experiments faster, smarter, and more resource-efficient.
Human experience and AI capabilities: a combined approach
While these successes are impressive, they also highlight an equally important reality. That means AI alone cannot replace the nuanced judgment and deep scientific intuition of human experts. Algorithms are excellent at generating large quantities of candidate materials and predicting performance metrics, but only experienced researchers can rigorously assess synthetic feasibility, physical and chemical principles, scalability to industrial quantities, safety considerations, and long-term environmental sustainability. The most effective AI-powered discovery teams are those in which domain specialists apply scientific intuition to filter AI-generated candidates, avoid costly dead ends, and steer research toward breakthrough innovation. This balance defines human-involved innovation, where AI accelerates and humans interpret, guide, and protect discoveries. ⁸ This symbiotic relationship between human insight and AI capabilities is critical to realizing the full potential of AI-driven materials discovery and ensuring that developments are translated into impactful and viable technologies.
education and training
If human judgment is to be at the heart of responsible AI, preparing the next generation with technical skills is essential. Economic demand for materials engineers with AI expertise is rapidly increasing. Energy, aerospace, electronics, and manufacturing companies are actively seeking engineers who can design experiments, interpret results, and use AI tools and machine learning to accelerate innovation. Incorporating AI literacy into materials science education is no longer an option, given that nearly all materials engineers of the future will engage in AI-enhanced, data-rich workflows. It’s essential. Universities are experimenting with hybrid curricula that integrate AI modules into core science and engineering courses. Hands-on capstone projects, crash courses, and “data bootcamps” are increasingly being used to teach not just what AI can do, but what AI should do. ⁹ By shaping researchers into innovators who can both harness and ask questions of AI, education ensures that technology serves humanity, rather than backfires.
Responsible AI for a prosperous future
Overall, AI is no longer a distant promise but a driving force in materials science. This shortens discovery timelines, enables sustainable design, and integrates manufacturing through digital twins and adaptive materials. However, important challenges remain, including data and prediction quality issues, the interpretability of AI models, and the limited number of researchers trained in AI. High-quality, standardized materials datasets remain in short supply, and many databases are incomplete, inconsistent, or limited, limiting robust model training and transferability. The complexity and context-sensitivity of materials, with performance often dependent on processing conditions and microstructure, further impedes generalization. Many ML models lack interpretability, reducing their reliability and making predictions difficult to integrate with physics-based frameworks. Additionally, the development and validation of AI platforms requires interdisciplinary expertise, which limits adoption. Equally important are the unresolved limitations of AI in understanding processing structure-property relationships and manufacturing feasibility. Most models optimize thermodynamic stability or target properties without considering processing, production constraints, cost, or supply chain instability, requiring human evaluation and adaptive feedback loops to match AI predictions with real-world feasibility. These barriers highlight why AI success ultimately depends on human expertise. But AI, guided by human creativity and responsibility, can unlock materials and technologies that are not only faster and smarter, but also transformative for society.
References
Nekuda Malik JA. The U.S. National Academies reports on Frontiers in Materials Research. MRS breaking news. 2019;44(5):329-334 Merchant, A. Butzner, S. Schoenholz, South Carolina. Eikor, M. Chong, G. Cubuk, ED Scaling deep learning for materials discovery. Nature 2023, 624 (7990), 80-85 Szymanski, NJ; Lendy, B. Fay, Y. Kumar, R.E.; He, T. Milstead, D. McDermott, Minnesota. Gallant, M. Kubek, ED; Merchant, A. et al. An autonomous laboratory to accelerate the synthesis of new materials. Nature 2023, 624 (7990), 86-91 Okabe, R., Cheng, M., Chotrattanapituk, A. et al. Integrating structural constraints in generative models for quantum materials discovery. nut. meter. (2025) Levine, DS et al. “Open Molecules 2025 (OMol25) dataset, evaluation, and models” arXiv preprint arXiv:2505.08762 (2025) Delgado-Licona, F., Alsaiari, A., Dickerson, H., et al. Power flow-driven data to accelerate autonomous inorganic materials discovery. Nat Chem Eng 2, 436–446 (2025) AE Siemenn et al. Self-supervised robotic system for autonomous contact-based spatial mapping of semiconductor properties. Science. Adv 11, eadw7071(2025) Ramprasad, R., Batra, R., Pilania, G. et al. Machine learning in materials informatics: Recent applications and prospects. npj Comput Mater 3, 54 (2017) TJ Oweida, A. Ul-Mahmood, MD Manning, S. Rigin, YG Yingling, “Combining materials and data science: Opportunities, challenges, and education in materials informatics,” MRS Advances 5 (2020) 1-18
This article will also be published in the quarterly magazine issue 24.
Source link
