LUO Di;

• 1:Department of Physics
• 2:Tsinghua University
• 3:Institute for Advanced Study
• 4:Tsinghua University

Abstract：

Artificial intelligence(AI)and physics are entering a new stage of deep interdisciplinary integration.Physics has long provided foundational ideas for AI,including symmetry,statistical mechanics,and quantum theory,while continuing to inspire new model architectures,learning principles,and computational paradigms.At the same time,as AI systems become increasingly complex,their internal mechanisms,collective behavior,and emergent dynamics are becoming new objects of interest and investigation in physics.In this Perspective,we introduce Physics-first AI,a research paradigm that regards AI as a complex informationprocessing system embedded in the physical world and seeks to understand,design,and build intelligent systems from fundamental physics principles and modes of reasoning.Physics-First AI aims to provide a common foundation for model architectures,training mechanisms,computational processes,and reasoning paradigms through a unified physics perspective.We trace the historical interplay between physics and AI and discuss how structure priors,statistical mechanics,quantum theory,and physics reasoning may together shape the next generation of intelligent systems.We further examine how this emerging paradigm could reshape physics education and the training of interdisciplinary researchers in the AI era.Physics-First AI offers a route toward a physics-grounded theory and design framework for next-generation intelligence,opening new opportunities for scientific AI,autonomous scientific discovery,and ultimately artificial general intelligence.

Key Words： Physics-first AI;artificial intelligence;symmetry;statistical mechanics;quantum theory;physics reasoning;physics intelligence model;quantum advantage

Abstract:

Keywords:

Foundation:

Authors: LUO Di;

DOI: 10.27024/j.wlygc.2026.07.08.is

References：

• [1]HOPFIELD J J.Neural networks and physical systems with emergent collective computational abilities[J].Proceedings of the National Academy of Sciences,1982,79(8):2554-2558．
• [2]FRISTON K.The free-energy principle:a unified brain theory?[J].Nature Reviews Neuroscience,2010,11(2):127-138．
• [3]COHEN T,WELLING M.Group equivariant convolutional networks[C]//International Conference on Machine Learning.New York City:PMLR,2016:2990-2999．
• [4]SOHL-DICKSTEIN J,WEISS E,MAHESWARANATHANN,et al.Deep unsupervised learning using nonequilibrium thermodynamics[C]//International Conference on Machine Learning.Lille:PMLR,2015:2256-2265．
• [5]MEHTA P,SCHWAB D J.An exact mapping between the variational renormalization group and deep learning[J/OL].arXiv,2014,arXiv:1410.3831.https：//arxiv.org/abs/1410.3831．
• [6]KAPLAN J,MCCANDLISH S,HENIGHAN T,et al.Scaling laws for neural language models[J/OL].arXiv,2020,arXiv:2001.08361.https：//arxiv.org/abs/2001.08361．
• [7]SCHR??DINGER E.What is life?The physical aspect of the living cell[M].Victoria,BC Canada:Rare Treasure Editions,2025．
• [8]LANDAUER R.Irreversibility and heat generation in the computing process[J].IBM Journal of Research and Development,1961,5(3):183-191．
• [9]FEYNMAN R P.Simulating physics with computers[M].Feynman and computation:exploring the limits of computers.USA:Perseus Books,1999:133-153．
• [10]SILVER D,HUANG A,MADDISON C J,et al.Mastering the game of Go with deep neural networks and tree search[J].Nature.2016,529(7587):484-489．
• [11]JUMPER J,EVANS R,PRITZEL A,et al.Highly accurate protein structure prediction with AlphaFold[J].Nature,2021,596(7873):583-589．
• [12]CARLEO G,CIRAC I,CRANMER K,et al.Machine learning and the physical sciences[J].Reviews of Modern Physics,2019,91(4):045002．
• [13]JIAO L,SONG X,YOU C,et al.AI meets physics:a comprehensive survey[J].Artificial Intelligence Review,2024,57(9):256．
• [14]BAHRI Y,KADMON J,PENNINGTON J,et al.Statistical mechanics of deep learning.Annual Review of Condensed Matter Physics,2020,11(1):501-528．
• [15]NOETHER E.Invariant variation problems[J].Transport Theory and Statistical Physics,1971,1(3):186-207．
• [16]BATZNER S,MUSAELIAN A,SUN L,et al.E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials[J].Nature Communications,2022,13(1):2453．
• [17]LUO D,CLARK B K.Backflow transformations via neural networks for quantum many-body wave functions[J].Physical Review Letters,2019,122(22):226401．
• [18]LUO D,CARLEO G,CLARK B K,et al.Gauge equivariant neural networks for quantum lattice gauge theories[J].Physical Review Letters,2021,127(27):276402．
• [19]KARNIADAKIS G E,KEVREKIDIS I G,LU L,et al.Physics-informed machine learning[J].Nature Reviews Physics,2021,3(6):422-440．
• [20]GREYDANUS S,DZAMBA M,YOSINSKI J.Hamiltonian neural networks[C]//Advances in Neural Information Processing Systems.2019:32．
• [21]CRANMER M,GREYDANUS S,HOYER S,et al.Lagrangian neural networks[J/OL].arXiv,2020,arXiv:2003.04630.https：//arxiv.org/abs/2003.04630．
• [22]MARKOVIC'D,MIZRAHI A,QUERLIOZ D,et al.Physics for neuromorphic computing[J].Nature Reviews Physics,2020,2(9):499-510．
• [23]LECUN Y.A path towards autonomous machine intelligence version 0.9.2,2022-06-27[EB/OL].Open Review,2022,62(1):1-62.https：//openreview.net/pdf?id=BZ5alrkVsf
• [24]CHEN Z,MCCARRAN J,VIZCAINO E,et al.TENG:Time-evolving natural gradient for solving PDEs with deep neural nets toward machine precision[C].Proceedings of the 41st International Conference on Machine Learning.Vienna,PMLR 235:7143-7162,2024．
• [25]BLEI D M,KUCUKELBIR A,MCAULIFFE J D.Variational inference:A review for statisticians[J].Journal of the American Statistical Association,2017,112(518):859-877．
• [26]LEVINE S.Reinforcement learning and control as probabilistic inference:Tutorial and review[J/OL].arXiv,2018,arXiv:1805.00909.https：//arxiv.org/abs/1805.00909．
• [27]OUYANG L,WU J,JIANG X,et al.Training language models to follow instructions with human feedback[C]//Advances in Neural Information Processing Systems.2022,35:27730-27744．
• [28]RAFAILOV R,SHARMA A,MITCHELL E,et al.Direct preference optimization:Your language model is secretly a reward model[C]//Advances in Neural Information Processing Systems.2023,36:53728-53741．
• [29]KWON C,AO P,THOULESS D J.Structure of stochastic dynamics near fixed points[J].Proceedings of the National Academy of Sciences.2005Sep 13,102(37):13029-33．
• [30]ZIYIN L,CHUANG I,GALANTI T,et al.Formation of representations in neural networks[C].InInternational Conference on Learning Representations.Vancouver,2025May 1(Vol.2025,pp.68535-68563）．
• [31]HABALA J,MARGOLIS G B,WANG T,et al.Agile asymmetric multi-legged locomotion:contact planning via geometric mechanics and spin model duality[J/OL].arXiv preprint arXiv:2602.09123.2026Feb 9．
• [32]HUANG H Y,CHOI S,MCCLEAN J R,et al.Vast World of Quantum Advantage[J].Physical Review X,2026,16(3):030501．
• [33]SCHULD M,KILLORAN N.Quantum machine learning in feature Hilbert spaces[J].Physical Review Letters,2019,122(4):040504．
• [34]GILY??N A,SU Y,LOW G H,et al.Quantum singular value transformation and beyond:exponential improvements for quantum matrix arithmetics[C]//Proceedings of the 51st Annual ACM SIGACT Symposium on Theory of Computing.2019:193-204．
• [35]MONTANARO A.Quantum speedup of Monte Carlo methods[J].Proceedings.Mathematical,Physical,and Engineering Sciences/The Royal Society,2015,471(2181):20150301．
• [36]YARKONI S,RAPONI E,B??CK T,et al.Quantum annealing for industry applications:Introduction and review[J].Reports on Progress in Physics,2022,85(10):104001．
• [37]BLEKOS K,BRAND D,CESCHINI A,et al.A review on quantum approximate optimization algorithm and its variants[J].Physics Reports,2024,1068:1-66．
• [38]GEORGESCU I M,ASHHAB S,NORI F.Quantum simulation[J].Reviews of Modern Physics,2014,86(1):153-185．
• [39]SHOR P W.Algorithms for quantum computation:discrete logarithms and factoring[C]//Proceedings of the 35th Annual Symposium on Foundations of Computer Science.Los Alamitos:IEEE,1994,124-134．
• [40]BIAMONTE J,WITTEK P,PANCOTTI N,et al.Quantum machine learning[J].Nature,2017,549(7671):195-202．
• [41]CEREZO M,VERDON G,HUANG H Y,et al.Challenges and opportunities in quantum machine learning[J].Nature Computational Science,2022,2(9):567-576．
• [42]STOUDENMIRE E,SCHWAB D.Supervised learning with tensor networks[C]//Advances in Neural Information Processing Systems 29.Red Hook:Curran Associates,2016．
• [43]CHEN Z,DANGOVSKI R,LOH C,et al.Quanta:Efficient high-rank fine-tuning of llms with quantum-informed tensor adaptation[C].Advances in Neural Information Processing Systems.Vancouver,2024Dec 16;37:92210-45．
• [44]CHEN Z,COMAS O,JIN Z,et al.L2M:Mutual Information Scaling Law for Long-Context Language Modeling[C].Advances in Neural Information Processing Systems.San Diego,2026Apr 23;38:138466-506．
• [45]HUANG H Y,BROUGHTON M,COTLER J,et al.Quantum advantage in learning from experiments[J].Science,2022,376(6598):1182-1186．
• [46]POLYA G.How to solve it:A new aspect of mathematical method[M].Princeton university press,1945Dec 31．
• [47]JEFFREYS H.The theory of probability[M].OuP Oxford,1998Aug 6．
• [48]SCHMIDT M,LIPSON H.Distilling free-form natural laws from experimental data[J].Science,2009,324(5923):81-85．
• [49]WU T,TEGMARK M.Toward an artificial intelligence physicist for unsupervised learning[J].Physical Review E,2019,100(3):033311．
• [50]LIU Z,TEGMARK M.Machine learning conservation laws from trajectories[J].Physical Review Letters,2021,126(18):180604．
• [51]FANG Y L,JIAN D S,LI X,et al.AI-Newton:A concept-driven physical law discovery system without prior physical knowledge[J/OL].arXiv,2025,arXiv:2504.01538.https：//arxiv.org/abs/2504.01538．
• [52]QIU S,DENG J,DENG Y,et al.PRBench:End-to-end paper reproduction in physics research[J/OL].arXiv,2026,arXiv:2603.27646.https：//arxiv.org/abs/2603.27646．
• [53]WANG W,HUANG D,LI J,et al.Cmphysbench:Abenchmark for evaluating large language models in condensed matter physics[C]//International Conference on Learning Representations 2026.Rio de Janeiro:ICLR,2026:3388-3421．
• [54]MIAO T,JIN W,ZHANG M,et al.PRL-Bench:A Comprehensive Benchmark Evaluating LLMs'Capabilities in Frontier Physics Research[J/OL].arXiv,2026,arXiv:2604.15411.https：//arxiv.org/abs/2604.15411．
• [55]PAN H,ROGGEVEEN J,BERG E,et al.CMT-benchmark:A benchmark for condensed matter theory built by expert researchers[C]//International Conference on Learning Representations 2026.Rio de Janeiro:ICLR,2026:35522-35540．
• [56]DENG K,WANG X,DUAN G,et al.Towards verifiable and self-correcting ai physicists for quantum many-body simulations[J/OL].arXiv,2026,arXiv:2604.00149.https：//arxiv.org/abs/2604.00149.