.png)
December 9, 2024
A recent study demonstrates the significant impact of Artificial Intelligence (AI) on the discovery of new materials. In a controlled experiment at a large US research laboratory with over 1,000 scientists, AI-assisted teams proved significantly more productive in the development of inorganic materials.
The research teams that worked with a specially developed AI tool discovered 44% more new materials than the control groups using conventional methods. Furthermore, the AI-supported teams filed 39% more patent applications, illustrating the direct impact of the technology on innovation. The novel materials find application in various fields, including healthcare, optics, and industrial manufacturing.
The AI system used is based on a combination of Graph Neural Networks and Reinforcement Learning. This technology was trained with data from extensive materials databases, including the "Materials Project" for crystal structures and the "Alexandria Materials Database" for molecular structures. The development process of the AI tool comprises three main phases:
1. Pre-training with known material structures.
2. Fine-tuning to specific application areas.
3. Reinforcement Learning based on experimental results to generate stable materials.
Researchers input the desired material properties into the neural network, which then suggests new structures that potentially exhibit these properties. The teams select the most promising structures, synthesize them, and subsequently test them in experiments and even in product prototypes. The results of these tests are fed back into the neural network to continuously improve its predictive capabilities.
Interestingly, the highest-performing researchers benefited most from the AI support. Scientists with lower performance, on the other hand, experienced only a small advantage. A possible explanation for this is that experienced researchers use their expertise to prioritize the most promising AI suggestions, while others expend resources on reviewing less promising suggestions.
A follow-up survey revealed that the job satisfaction of the researchers using the AI tool tended to be lower. This could be due to the tool taking over some of the more creative steps in the research process. The scientists primarily had to choose which of the suggested materials should be transferred to the next phase of development.
The study underscores the potential of AI to significantly accelerate materials research. At the same time, it becomes clear that human expertise remains essential to evaluate the AI suggestions and effectively manage the research process. The results indicate that the combination of AI and human expertise is the key to future innovations in materials science.
Bibliographie: https://medium.com/@frederikbonde/ai-in-materials-science-human-expertise-drives-44-more-discoveries-6504d2e620b6 https://ground.news/article/a-randomized-study-at-a-corporate-lab-employing-more-than-1-000-researchers-teams-using-ai-discovered-44-more-new-materials-than-teams-with-s https://www.paterhn.ai/impact-and-insights/ai-in-materials-science-how-human-expertise-drives-44-more-discoveries https://www.linkedin.com/posts/william-marcellino-ph-d-41982a19_ai-generativeai-productivity-activity-7261723406788710400-ICOw https://www.nature.com/articles/d41586-024-04025-6 https://deepmind.google/discover/blog/millions-of-new-materials-discovered-with-deep-learning/ https://www.mckinsey.de/~/media/mckinsey/locations/europe%20and%20middle%20east/deutschland/news/presse/2024/2024%20-%2005%20-%2023%20mgi%20genai%20future%20of%20work/mgi%20report_a-new-future-of-work-the-race-to-deploy-ai.pdf https://www.imeche.org/news/news-article/mit-team-discovers-tough-and-durable-new-materials-using-ai https://www.sciencedirect.com/science/article/pii/S0304405X2300185X https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2023/11/Microsoft_Accelerating-Sustainability-with-AI-A-Playbook-1.pdf