arxiv: 2604.14718 · v1 · submitted 2026-04-16 · 💻 cs.AI · cond-mat.dis-nn· hep-th

Recognition: unknown

The Agentification of Scientific Research: A Physicist's Perspective

Xiao-Liang Qi

Authors on Pith no claims yet

Pith reviewed 2026-05-10 11:09 UTC · model grok-4.3

classification 💻 cs.AI cond-mat.dis-nnhep-th

keywords AI for sciencelarge language modelsscientific collaborationagentificationcontinuous learningscientific publishingresearch evaluation

0 comments

The pith

AI's core impact on science is a shift in how knowledge is carried and shared, making AI a collaborator rather than a tool

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper claims that the rise of large language models represents more than automation; it changes the basic mechanisms for transmitting and replicating complex information and expertise. This matters for science because it could alter the organization of collaboration among researchers, the way discoveries are made, how findings are published, and how they are evaluated. The author maps out a step-by-step process in which AI begins as a supportive tool and progresses to functioning as an equal partner in scientific work. To achieve original contributions, these AI systems must support ongoing learning and a variety of perspectives.

Core claim

The most important significance of the AI revolution, especially the rise of large language models, lies not simply in automation, but in a fundamental change in how complex information and human know-how are carried, replicated, and shared. From this perspective, AI for Science is especially important because it may transform not only the efficiency of research, but also the structure of scientific collaboration, discovery, publishing, and evaluation. The article outlines a gradual path from AI as a research tool to AI as a scientific collaborator, and discusses how AI is likely to fundamentally reshape scientific publication. It also argues that continuous learning and diversity of ideas 0

What carries the argument

Agentification of research, the process turning AI into scientific collaborators that carry and replicate know-how

If this is right

The structure of scientific collaboration will incorporate AI agents as active participants.
Scientific publishing will be fundamentally reshaped to account for AI involvement in content creation and review.
Research evaluation methods will evolve to assess contributions from both humans and AI systems.
Original scientific discovery will depend on AI maintaining continuous learning and idea diversity.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Researchers may develop new practices for interacting with AI to maximize collaborative output.
Fields could see faster integration of knowledge across disciplines through AI's ability to replicate diverse expertise.
Pilot projects using AI agents in controlled research settings could verify their capacity for independent idea generation.

Load-bearing premise

That AI systems can acquire continuous learning abilities and sustain diversity of ideas to make original discoveries, with the shift to collaborator status occurring without major barriers.

What would settle it

Evidence that AI systems, despite extensive data exposure, repeatedly fail to produce or validate any novel, verifiable scientific insights without constant human guidance at key steps.

Figures

Figures reproduced from arXiv: 2604.14718 by Xiao-Liang Qi.

**Figure 1.** Figure 1: Illustration of the three major transformations of information dynamics in Earth’s history: life, human language, and the AI revolution. generations without waiting for biological inheritance. Compared with genetic evolution, linguistic and cultural evolution proceeded at a dramatically faster pace. Human societies could accumulate ideas, institutions, and technologies through communication, education, and… view at source ↗

**Figure 2.** Figure 2: Illustration of the major pain points in scientific research, including the time cost of understanding prior work, the loss of tacit knowledge, limits of collaboration, and administrative burden. To understand what AI brings to research, we must review common problems currently facing scientific enquiry. While challenges vary by field, several are universal: 1. Time Costs: Understanding industry progress a… view at source ↗

**Figure 3.** Figure 3: Illustration of the agentification of scientific research, from AI use of research tools and automation of repetitive work to scientific collaboration, cross-disciplinary interaction, and agentic publishing. The application of LLMs in science is already underway, with AI agents assisting research in fields such as biology, mathematics, chemistry, theoretical physics, and machine learning. Although current… view at source ↗

**Figure 4.** Figure 4: Illustration of why the next step for AI in science is real-time learning and diversity of ideas, enabling continuous adaptation to frontier research and more original scientific discovery. 2.3 Challenges in AI for Science The opportunities described above are substantial, but they should not be confused with fully realized capabilities. To move from promising demonstrations to a genuine transformation of… view at source ↗

read the original abstract

This article argues that the most important significance of the AI revolution, especially the rise of large language models, lies not simply in automation, but in a fundamental change in how complex information and human know-how are carried, replicated, and shared. From this perspective, AI for Science is especially important because it may transform not only the efficiency of research, but also the structure of scientific collaboration, discovery, publishing, and evaluation. The article outlines a gradual path from AI as a research tool to AI as a scientific collaborator, and discusses how AI is likely to fundamentally reshape scientific publication. It also argues that continuous learning and diversity of ideas are essential if AI is to play a meaningful role in original scientific discovery.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Qi argues AI mainly changes how scientific know-how gets replicated and shared, potentially restructuring research, but the piece stays at the level of plausible speculation.

read the letter

The main point here is that the real significance of LLMs and similar systems is not automation alone but a shift in carrying and replicating complex expertise, which could then alter collaboration, discovery, publishing, and evaluation in science. Qi frames this as a gradual move from AI as tool to AI as collaborator, with continuous learning and idea diversity as prerequisites for any original contributions from the systems themselves. That framing is the clearest new angle, presented from a physicist's practical standpoint rather than from an AI lab or policy angle. It pulls together existing trends without claiming immediate breakthroughs, which keeps the discussion grounded in observable patterns like how tools have changed science before. The writing is direct and avoids hype about timelines. The soft spots are straightforward: this is argument and extrapolation only, with no data, case studies, simulations, or formal models to support the structural predictions. The central assumption that AI will reliably develop ongoing learning and sustain idea diversity enough to matter for discovery is stated but not tested against current limitations or counterexamples. No technical barriers or social pushback get much attention either. Readers who track AI-for-science conversations will find this useful as a prompt for thinking about credit, evaluation, and lab organization. It does not deliver new methods or results, so it is not something most people would cite directly. The thinking is coherent and engages the literature honestly, so the paper deserves a serious referee to pressure-test the assumptions and suggest where more evidence could be added.

Referee Report

0 major / 2 minor

Summary. The manuscript is a perspective article arguing that the primary significance of the AI revolution, especially large language models, is not automation but a fundamental shift in how complex information and human know-how are carried, replicated, and shared. It claims this will transform the structure of scientific collaboration, discovery, publishing, and evaluation, outlining a gradual path from AI as a research tool to AI as a collaborator. The paper emphasizes that continuous learning and diversity of ideas are essential for AI to contribute to original scientific discovery.

Significance. If the perspective holds, it offers a timely interpretive framework for physicists and AI researchers on the structural implications of AI for science, moving beyond efficiency gains to changes in knowledge replication and institutional practices. The argument draws on historical patterns and current trends to highlight potential shifts in collaboration and evaluation, providing a coherent narrative that could inform discussions on AI for Science.

minor comments (2)

The transition from tool to collaborator is described qualitatively; adding a brief timeline or milestone examples in the relevant section would strengthen readability without altering the perspective nature.
The abstract and introduction both state the core thesis on know-how replication; consider consolidating to avoid minor repetition.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive assessment of the manuscript and for recommending acceptance. The referee's summary accurately captures the central thesis that the significance of AI, particularly large language models, lies in reshaping how complex information and expertise are replicated and shared, with implications for scientific collaboration, discovery, publishing, and evaluation.

Circularity Check

0 steps flagged

No significant circularity in perspective article

full rationale

The manuscript is a perspective article advancing interpretive opinions on AI's impact on scientific processes. It contains no formal derivation chain, equations, quantitative predictions, or fitted parameters. Claims rest on general historical observations and forward-looking speculation without reducing any result to self-defined inputs, self-citations as load-bearing premises, or renaming of known results. The central argument about AI transforming know-how replication is presented as opinion, not a derived proposition requiring validation against its own premises.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on assumptions about the future trajectory of AI capabilities and the requirements for original discovery, without introducing new parameters, entities, or formal axioms beyond domain-level expectations about AI development.

axioms (2)

domain assumption AI can progress from tool to collaborator through gradual development.
Invoked in the outlined path from current AI use to future integration in scientific work.
domain assumption Continuous learning and diversity of ideas are required for AI to enable original discovery.
Presented as essential conditions in the discussion of meaningful AI roles in science.

pith-pipeline@v0.9.0 · 5409 in / 1295 out tokens · 49596 ms · 2026-05-10T11:09:15.822922+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

36 extracted references · 27 canonical work pages · 4 internal anchors

[1]

Bubeck, C

Sébastien Bubeck, Christian Coester, Ronen Eldan, et al. Early science acceleration experi- ments with GPT-5, 2025. URL https://arxiv.org/abs/2511.16072

work page arXiv 2025
[2]

(2026, February 2)

Andres M. Bran, Sam Cox, Oliver Schilter, et al. Augmenting large language models with chemistry tools. Nature Machine Intelligence , 6(5):525–535, 2024. doi: 10.1038/s42256-024- 00832-8. URL https://doi.org/10.1038/s42256-024-00832-8

work page doi:10.1038/s42256-024- 2024
[3]

Accelerating clinical evidence synthesis with large language models

Zifeng Wang, Lang Cao, Benjamin Danek, et al. Accelerating clinical evidence synthesis with large language models. npj Digital Medicine , 8:509, 2025. doi: 10.1038/s41746-025-01840-7. URL https://doi.org/10.1038/s41746-025-01840-7

work page doi:10.1038/s41746-025-01840-7 2025
[4]

Li, Emily B

Michael Y. Li, Emily B. Fox, and Noah D. Goodman. Automated statistical model discovery with language models, 2024. URL https://arxiv.org/abs/2402.17879

work page arXiv 2024
[5]

Single-minus graviton tree am- plitudes are nonzero, 2026

Alfredo Guevara, Alexandru Lupsasca, David Skinner, et al. Single-minus graviton tree am- plitudes are nonzero, 2026. URL https://cdn.openai.com/pdf/graviton.pdf. OpenAI preprint PDF

2026
[6]

Brenner, Vincent Cohen-Addad, and David Woodruff.Solving an Open Problem in Theoretical Physics using AI-Assisted Discovery

Michael P. Brenner, Vincent Cohen-Addad, and David Woodruff. Solving an open problem in theoretical physics using AI-assisted discovery, 2026. URL https://arxiv.org/abs/2603.04735

work page arXiv 2026
[7]

org/abs/2506.06214

Sirui Lu, Zhijing Jin, Terry Jingchen Zhang, et al. Can theoretical physics research benefit from language agents?, 2025. URL https://arxiv.org/abs/2506.06214

work page arXiv 2025
[8]

Agent laboratory: Using LLM agents as research assistants

Samuel Schmidgall, Yusheng Su, Ze Wang, et al. Agent laboratory: Using LLM agents as research assistants. In Findings of the Association for Computational Linguistics: EMNLP 2025, 2025. URL https://aclanthology.org/2025.findings-emnlp.320/

2025
[9]

SciSciGPT: Advancing human-AI collaboration in the science of science

Erzhuo Shao, Yifang Wang, Yifan Qian, et al. SciSciGPT: Advancing human-AI collaboration in the science of science. Nature Computational Science, 2025. doi: 10.1038/s43588-025-00906-

work page doi:10.1038/s43588-025-00906- 2025
[10]

URL https://doi.org/10.1038/s43588-025-00906-6

work page doi:10.1038/s43588-025-00906-6
[11]

From Paper to Program: Accelerating Quantum Many-Body Algorithm Development via a Multi-Stage LLM-Assisted Workflow

Yi Zhou. From paper to program: A multi-stage LLM-assisted workflow for accelerating quantum many-body algorithm development, 2026. URL https://arxiv.org/abs/2604.04089

work page internal anchor Pith review Pith/arXiv arXiv 2026
[13]

V ASPilot: MCP-Facilitated Multi-Agent Intelligence for Autonomous V ASP Simulations, August 2025

Jiaxuan Liu, Tiannian Zhu, Caiyuan Ye, et al. V ASPilot: MCP-facilitated multi-agent intelli- gence for autonomous V ASP simulations, 2025. URL https://arxiv.org/abs/2508.07035

work page arXiv 2025
[14]

Materialsgalaxy: A plat- form fusing experimental and theoretical data in condensed matter physics

Tiannian Zhu, Zhong Fang, Quansheng Wu, and Hongming Weng. Materialsgalaxy: A plat- form fusing experimental and theoretical data in condensed matter physics. Chinese Physics B, 34(12):120702, 2025

2025
[15]

Towards an AI co-scientist

Juraj Gottweis, Wei-Hung Weng, Alexander Daryin, Tao Tu, Anil Palepu, Petar Sirkovic, et al. Towards an AI co-scientist, 2025. URL https://arxiv.org/abs/2502.18864. 12

work page internal anchor Pith review arXiv 2025
[16]

Bohrium +

Linfeng Zhang, Siheng Chen, Yuzhu Cai, et al. Bohrium + SciMaster: Building the infrastruc- ture and ecosystem for agentic science at scale, 2025. URL https://arxiv.org/abs/2512.20469

work page arXiv 2025
[17]

Scimaster: Towards general-purpose scientific ai agents, part i

Jingyi Chai, Shuo Tang, Rui Ye, Yuwen Du, Xinyu Zhu, Mengcheng Zhou, Yanfeng Wang, Yuzhi Zhang, Linfeng Zhang, Siheng Chen, et al. Scimaster: Towards general-purpose scientific ai agents, part i. x-master as foundation: Can we lead on humanity’s last exam? arXiv preprint arXiv:2507.05241, 2025

work page arXiv 2025
[18]

The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery

Chris Lu, Cong Lu, Robert Tjarko Lange, et al. The AI scientist: Towards fully automated open-ended scientific discovery, 2024. URL https://arxiv.org/abs/2408.06292

work page internal anchor Pith review arXiv 2024
[19]

Exploring the use of AI authors and reviewers at Agents4Science

Federico Bianchi, Owen Queen, Nitya Thakkar, Eric Sun, James Zou, et al. Exploring the use of AI authors and reviewers at Agents4Science. Nature Biotechnology, 44:11–14, 2026. doi: 10.1038/s41587-025-02963-8. URL https://doi.org/10.1038/s41587-025-02963-8

work page doi:10.1038/s41587-025-02963-8 2026
[20]

Generative AI in scientific publishing: Disruptive or destructive? Nature Reviews Urology , 21:1–2, 2024

Riccardo Bertolo and Alessandro Antonelli. Generative AI in scientific publishing: Disruptive or destructive? Nature Reviews Urology , 21:1–2, 2024. doi: 10.1038/s41585-023-00836-w. URL https://doi.org/10.1038/s41585-023-00836-w

work page doi:10.1038/s41585-023-00836-w 2024
[21]

Scientific production in the era of large language models.Science, 390(6779):1240–1243, 2025

Keigo Kusumegi, Xinyu Yang, Paul Ginsparg, et al. Scientific production in the era of large language models. Science, 390(6779):1240–1243, 2025. doi: 10.1126/science.adw3000. URL https://doi.org/10.1126/science.adw3000

work page doi:10.1126/science.adw3000 2025
[22]

Quantifying large language model usage in scientific papers

Weixin Liang, Yaohui Zhang, Zhengxuan Wu, et al. Quantifying large language model usage in scientific papers. Nature Human Behaviour , 9:2599–2609, 2025. doi: 10.1038/s41562-025- 02273-8. URL https://doi.org/10.1038/s41562-025-02273-8

work page doi:10.1038/s41562-025- 2025
[23]

Model context protocol, 2024

Anthropic. Model context protocol, 2024. URL https://modelcontextprotocol.io/docs/getting- started/intro

2024
[24]

Agent skills protocol, 2025

Anthropic. Agent skills protocol, 2025. URL https://agentskills.io/home

2025
[25]

arXiv:2503.13517 , year =

Hao Cui, Zahra Shamsi, Gowoon Cheon, et al. CURIE: Evaluating LLMs on multitask scientific long context understanding and reasoning, 2025. URL https://arxiv.org/abs/2503.13517

work page arXiv 2025
[26]

Roggeveen, Erez Berg, et al

Haining Pan, James V. Roggeveen, Erez Berg, et al. CMT-benchmark: A benchmark for condensed matter theory built by expert researchers, 2025. URL https://arxiv.org/abs/2510 .05228

2025
[27]

Expert evaluation of LLM world models: A high- 𝑡𝑐 superconductivity case study, 2025

Haoyu Guo, Maria Tikhanovskaya, Paul Raccuglia, et al. Expert evaluation of LLM world models: A high- 𝑡𝑐 superconductivity case study, 2025. URL https://arxiv.org/abs/2511.03782

work page arXiv 2025
[28]

Qmbench: A research level benchmark for quantum materials research

Yanzhen Wang, Yiyang Jiang, Diana Golovanova, Kamal Das, Hyeonhu Bae, Yufei Zhao, Huu- Thong Le, Abhinava Chatterjee, Yunzhe Liu, Chao-Xing Liu, et al. Qmbench: A research level benchmark for quantum materials research. arXiv preprint arXiv:2512.19753 , 2025

work page arXiv 2025
[29]

Cmphysbench: A benchmark for evaluating large language models in condensed matter physics.arXiv preprint arXiv:2508.18124, 2025

Weida Wang, Dongchen Huang, Jiatong Li, Tengchao Yang, Ziyang Zheng, Di Zhang, Dong Han, Benteng Chen, Binzhao Luo, Zhiyu Liu, et al. Cmphysbench: A benchmark for evaluating large language models in condensed matter physics. arXiv preprint arXiv:2508.18124 , 2025

work page arXiv 2025
[30]

Towards verifiable and self-correcting ai physicists for quantum many-body simulations

Ken Deng, Xiangfei Wang, Guijing Duan, Chen Mo, Junkun Huang, Runqing Zhang, Ling Qian, Zhiguo Huang, Jize Han, and Di Luo. Towards verifiable and self-correcting ai physicists for quantum many-body simulations. arXiv preprint arXiv:2604.00149 , 2026. 13

work page internal anchor Pith review arXiv 2026
[31]

Continual learning for large language models: A survey.arXiv:2402.01364, 2024

Tongtong Wu, Linhao Luo, Yuan-Fang Li, et al. Continual learning for large language models: A survey, 2024. URL https://arxiv.org/abs/2402.01364

work page arXiv 2024
[32]

A large-scale comparison of divergent creativity in humans and large language models

Dawei Wang, Difang Huang, Haipeng Shen, and Brian Uzzi. A large-scale comparison of divergent creativity in humans and large language models. Nature Human Behaviour , 2025. doi: 10.1038/s41562-025-02331-1. URL https://doi.org/10.1038/s41562-025-02331-1

work page doi:10.1038/s41562-025-02331-1 2025
[33]

doi:10.1038/s4 1586-025-09527-5

Qianyue Hao, Fengli Xu, Yong Li, James Evans, et al. Artificial intelligence tools expand scientists’ impact but contract science’s focus. Nature, 649:1237–1243, 2026. doi: 10.1038/s4 1586-025-09922-y. URL https://doi.org/10.1038/s41586-025-09922-y

work page doi:10.1038/s4 2026
[34]

Time, information and artificial intelligence

Xiao-Liang Qi. Time, information and artificial intelligence. Physics, 2024. doi: 10.7693/wl 20240601. URL https://wuli.iphy.ac.cn/cn/article/doi/10.7693/wl20240601 . Chinese article; page title also gives the English title “Time, information and artificial intelligence”

work page doi:10.7693/wl 2024
[35]

Teaching and mentoring the ai scientists, April 2025

Xiao-Liang Qi. Teaching and mentoring the ai scientists, April 2025. URL https://pirsa.org/ 25040066. PIRSA:25040066

2025
[36]

Teaching and mentoring the ai scientists

Xiao-Liang Qi. Teaching and mentoring the ai scientists. YouTube video, October 2025. URL https://www.youtube.com/watch?v=vYkYT1aBlVo . Title inferred from the corresponding PIRSA lecture link supplied by the author

2025
[37]

A brief perspective on the artificial intelligence revolution

Xiao-Liang Qi. A brief perspective on the artificial intelligence revolution. ai4.science discussion forum post, January 2026. URL https://forum.ai4.science/t/a-brief-perspective-on-the- artificial-intelligence-revolution/65. Posted January 19, 2026. 14

2026