AI for research (AI4R) is a significant change in research methods. The scientific and technological circles should not only pay attention to "AI for Science" (AI4S), but also attach great importance to "AI for Technology" (AI4T); Not only should we focus on the Large Language Model (LLM), but we should also pay more attention to the Large Science Model (LSM). The breakthrough of artificial intelligence mainly relies not on large computing power, but on the transformation of computational models. China should strive to make disruptive innovations on foundation models. AI4R is suitable for combinatorial search of complex problems, and neural network models may be close to the complexity threshold point that can handle difficult problems. One trend in AI4R is to abandon absoluteness, embrace uncertainty, and we should tolerate "black-box models" appropriately for a certain period of time.
[1] 李国杰. 智能化科研（AI4R）：第五科研范式[J]. 中国科学院院刊, 2024, 39(1): 1-9.
[2] Jumper J, Evans R, Pritzel A, et al. Highly accurate protein structure prediction with AlphaFold[J]. Nature, 2021, 596: 583-589.
[3] Guo Z Q, Lu D H, Yan Y J, et al. Extending the limit of molecular dynamics with ab initio accuracy to 10 billion atoms[C]. Seoul: Proceedings of the 27th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, 2022: 205-218.
[4] Cheng S Y, Jin P W, Guo Q, et al. Pushing the limits of machine design: Automated CPU design with AI[J/OL]. [2023-12-19]. https://arxiv.org/abs/2306.12456.
[5] Lam R, Sanchez-Gonzalez A, Willson M, et al. Learning skillful medium-range global weather forecasting[J]. Science, 2023, 382(6677): 1416-1421.
[6] Merchant A, Batzner S, Schoenholz S S, et al. Scaling deep learning for materials discovery[J]. Nature, 2023, 624: 80-85.
[7] Copeland B J. The essential turing[M]. Oxford: Clarendon Press, 2004: 411-432.
[8] John von Neumann. Theory of self-reproducing automata [M]. Urbana and London: University of Illinois Press, 2002.
