TLDR: This paper, from an NSF workshop, outlines a strategic vision for integrating Artificial Intelligence (AI) with the Mathematical and Physical Sciences (MPS). It highlights a ‘two-way street’ where AI accelerates scientific discovery and MPS insights enhance AI development. Key priorities include fostering interdisciplinary research, building scalable AI infrastructure, cultivating specific AI techniques like simulation-based inference and foundation models, and developing an AI-literate workforce. The document details opportunities across astronomy, chemistry, materials research, mathematics, and physics, emphasizing collaboration and the ‘Science of AI’ to drive transformative breakthroughs.
A recent community paper, stemming from the NSF Future of AI+MPS Workshop held in March 2025, offers a comprehensive look into the evolving relationship between Artificial Intelligence (AI) and the Mathematical and Physical Sciences (MPS). This insightful document, titled “The Future of Artificial Intelligence and the Mathematical and Physical Sciences (AI+MPS) Community Paper from the NSF Future of AI+MPS Workshop,” was organized by a team including Andrew Ferguson, Marisa LaFleur, Lars Ruthotto, Jesse Thaler, Yuan-Sen Ting, Pratyush Tiwary, and Soledad Villar. It highlights a crucial moment to strengthen the bond between AI and science, aiming to proactively use AI for scientific discovery and enhance AI development through fundamental scientific concepts.
The paper emphasizes a “two-way street” of mutual innovation. On one hand, AI is a powerful tool accelerating scientific discovery across MPS domains like astronomy, chemistry, materials research, mathematics, and physics. On the other, insights from MPS are vital for improving AI innovation and understanding, pushing the boundaries of what AI can achieve. This synergy is already evident, with the 2024 Nobel Prizes in Physics and Chemistry recognizing foundational AI methods and their applications in protein design.
Strategic Vision for AI+MPS
The workshop identified three strategic priorities to guide the future of AI+MPS:
-
Enable AI+MPS Research in Both Directions: This involves leveraging AI to advance scientific discovery and using scientific principles to deepen our understanding of AI.
-
Build an Interdisciplinary AI+MPS Community: Fostering collaboration across disciplines and with AI experts is key to sharing knowledge and amplifying research impact.
-
Foster AI+MPS Education and Workforce Development: Equipping current and future researchers with AI literacy is essential for continued progress in both science and AI.
Key Opportunities for Growth
The paper outlines several key opportunities for funding agencies, universities, and individual researchers to build the future of AI+MPS. These include advocating for diverse funding streams to support interdisciplinary research at various scales, from large institutes to individual projects and industry collaborations. A significant focus is placed on pursuing the “Science of AI,” which involves applying scientific insights to develop more robust, interpretable, and accurate AI tools. This means creating new AI architectures, improving computational efficiency, and building physics-based AI simulations.
Establishing scalable AI infrastructures is also critical, encompassing computing resources, data management, and the creation of standardized benchmarks for AI+MPS. The document stresses the need for accessible cloud computing, sustained software development, and curated data archives that are AI-ready and publicly accessible. Facilitating cross-disciplinary collaborations through workshops, conferences, and knowledge transfer initiatives is also highlighted as essential.
Furthermore, the paper identifies key AI techniques for science that have broad applicability across MPS domains. These include Simulation-Based Inference (SBI) for parameter estimation, multi-scale simulations that bridge different physical phenomena, and robust Uncertainty Quantification (UQ) to ensure reliable predictions. The development of domain-informed foundation models, AI for experimental control (leading to “self-driving labs”), and data-efficient methods for small datasets are also crucial areas for advancement.
AI is also seen as a tool for streamlining research itself, acting as an “AI co-pilot” for literature search and hypothesis generation, and enabling “digital twins” for complex systems. To support this, a strong emphasis is placed on educating and training an AI+MPS workforce across all career stages, from faculty upskilling to undergraduate and K-12 education. This includes developing interdisciplinary PhD programs and graduate certificates, and fostering industry partnerships.
Also Read:
- AI’s Scientific Endeavor: The Indispensable Role of Verification
- Enhancing AI Materials Discovery with Physics-Guided Reasoning
Domain-Specific Advancements
The paper delves into how AI is specifically impacting and being impacted by each MPS domain:
-
Astronomical Sciences (AST): AI is crucial for processing massive datasets from new observatories, advancing parameter inference, anomaly detection, and optimizing telescope operations. AST also provides rich, open datasets and physics-informed frameworks that challenge and advance AI.
-
Chemistry (CHE): AI is revolutionizing reaction prediction, molecular design, and accelerating simulations. Chemistry, in turn, has inspired AI innovations like diffusion models and sparse machine learning techniques.
-
Materials Research (DMR): AI accelerates molecular and materials discovery, design, and characterization, particularly in the small-data regime. DMR contributes to AI through symmetry-aware networks and the development of self-driving labs.
-
Mathematical Sciences (DMS): Mathematics and statistics form the foundational backbone of AI, providing essential techniques for understanding and improving AI models. AI is also advancing mathematical discovery, scientific computing, and theorem proving.
-
Physics (PHY): AI is used to accelerate theoretical calculations, improve experimental operations, and analyze unique datasets. Physics advances AI through developments in real-time AI, equivariant neural networks, and understanding neural scaling laws.
The paper concludes by urging funding agencies, educational institutions, and individual researchers to embrace an intentional strategy to build a dynamic AI+MPS community. This collaborative effort promises to offer profound insights into AI, accelerate the pace of scientific discovery, and develop robust tools for both science and AI. For more details, you can read the full paper here.
