Stronger artificial intelligence breakthroughs hasten their own exhaustion
Authors: Beichen Lu, Honglin Bao, Qi Huang, Qingguo Meng, James A. Evans
Organizations: School of Public Policy and Management, Tsinghua University, Beijing, China. · Knowledge Lab, University of Chicago, Chicago, IL, USA. · Data Science Institute, University of Chicago, Chicago, IL, USA. · Department of Sociology, University of Chicago, Chicago, IL, USA. · Santa Fe Institute, Santa Fe, NM, USA.
Progress in artificial intelligence is widely assumed to scale with effort: more researchers, papers and investment should accelerate discovery. Theories of technological growth instead locate the difficulty of discovery in the space of ideas. Here we show that neither scale nor space is enough: breakthroughs depend on diverse attention, which collective concentration can deplete. Across 36 years of computer science, we analyze 7.2 million papers, 73,089 peer reviews and 3.7 million researchers and their topic shifts. The field expanded while its attention narrowed sharply after the Transformer. Over half of topic-shifting researchers moved into AI, especially mid-career ones, and reviewer confidence fell by a full point on a 5-point scale. Each successive breakthrough, from AlexNet to large language models, raised the field's carrying capacity for ideas but was followed by steeper exhaustion of those ideas. Bigger breakthroughs capture attention more completely, depleting the diversity their successors may draw upon. Structural biology shows the same cycle after cryo-EM and AlphaFold. Simulations show how well-intentioned and widely used policies such as reciprocal reviewing empower prolific authors and further narrow what the field accepts. We call for institutions that protect diverse attention in artificial intelligence rather than reward concentration on the winning paradigm.
Figures & tables
Figure 1: Attention concentrates. a , b , Yearly Gini coefficients of research attention ( a ) and citations ( b ) hold flat then break upward in 2017. c , AI’s share of computer-science publications increases between 2017 and 2025, while systems and theory contract. d , Author migration probability into AI peaks at age 31.8 and the 82nd yearly publication-rank percentile. e , Across 73,089 peer reviews from 2017–2025, reviewer confidence declines from 3.9 to 2.9 on a 0–5 scale (left). Toxic comments increased after 2017, were briefly tempered by the NeurIPS move to OpenReview, and then rebounded (right). Lines in a , b , d , and e are fitted using a three-year rolling average.
Figure 2: Cyclical exhaustion and policy backfire. a , Each unique noun in the stable corpus is treated as an idea; ideas are born, die after three years of disuse, and may reactivate. b , The yearly share of newly born ideas rises sharply at AlexNet and the Transformer and then declines. c , Breakthroughs not only birth ideas; they bury them with a lag. d , Citation share for the original AlexNet and Transformer papers peaks and declines for both. Lines in b – d are fitted using a three-year rolling average. e , Active-idea populations fit logistic carrying-capacity curves across three successive cycles (solid lines), with the dashed tails marking observed deviation below each modeled ceiling. The drop is modest after AlexNet, severe after the Transformer, and emerges before the new ceiling under LLMs. The cycle is compressing. f , Existing policies may not address this: greater reciprocity strength ρ (high-submission reviewers being asked to review even more) drives stronger convergence toward the tastes of high-submission established researchers.
Appendix figures & tables6 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 1: Robust results on attention concentration. Gini coefficient under different TF-IDF keyword extraction parameters. Panels A, B, and C correspond to k=5 , k=10 , and k=15 , respectively. Lines are fitted using a three-year rolling average.
Figure 2: Robust results on idea representation. Robustness of idea lifecycle metrics across idea representation methods. Conclusions are robust.
Figure 3: Illustration of the four-stage spiral of technological revolution. Across 36 years of AI development including three major technological advances (AlexNet, the Transformer, and LLMs), we repeatedly observe the same pattern, a four-stage spiral of technological revolution : initiation , where a new paradigm has just emerged; growth , where the S-curve reaches its steepest slope and new ideas proliferate; saturation , where the idea population calcifies and may even deteriorate; and finally breakthrough , a largely unpredictable transformation that lifts the carrying capacity and restarts the cycle.
Figure 4: Cyclical exhaustion in structural biology. Left: Recognition of cryo-EM came later than the peak in new ideas. AlphaFold produced a second, smaller peak. Right: Active ideas follow the same growth pattern as in computer science: each breakthrough raises the carrying capacity of idea space, which is then repeatedly exhausted.
(1)
(2)
Logit
Logit
Age
0.025***
0.073***
(0.000)
(0.001)
Age 2
-0.001***
(0.000)
Prior Status
0.265***
1.083***
Appendix
Table 1: Determinants of author migration to AI (using yearly publication percentile to represent prior status)
(1)
(2)
Logit
Logit
Age
0.021***
0.056***
(0.000)
(0.001)
Age 2
-0.001***
(0.000)
Prior Status
0.395***
0.490***
Appendix
Table 2: Determinants of author migration to AI (using yearly citation percentile to represent prior status)
Do research topics in artificial intelligence grow gradually, or do they advance through abrupt, detectable jumps? Analyzing 80,814 accepted main-track papers from five premier AI conferences (ACL, CVPR, ICLR, ICML, NeurIPS) spanning 2017 to 2025, we show major AI topics advance through topical phase transitions: remaining marginal for years, then surging across venues within one to three years. Large language models became the dominant cross-venue topic by 2025, diffusion models rose with comparable abruptness, and language-model methods crossed into computer vision via vision-language models, whereas reinforcement learning compounded smoothly, distinguishing genuine phase transitions from ordinary growth. This structure is our primary contribution: a large-scale, cross-venue characterization of how AI research reorganizes. We then ask whether a transition leaves a detectable footprint before it peaks. We define an early-warning signature, four publication-dynamics criteria frozen on 2017-2021 data, and evaluate it out of sample on 2023-2025 transitions, obtaining a precision of 27% and recall of 63% against a 13.5% base rate. Applied to 2025 data, the signature flags reasoning and test-time compute, agentic AI, multimodal LLMs, retrieval-augmented generation, and world models as topics to monitor over 2026-2028. The source code is also publicly available on GitHub at https://github.com/KurbanIntelligenceLab/ai-phase-transitions.
Rasul Khanbayov, Hasan Kurban
College of Science and Engineering, Hamad Bin Khalifa University, Doha, Qatar.
Generative artificial intelligence raises a central question for scientific training and organization. Is research shifting from deep specialization toward broad individual knowledge? We examine this proposition across papers, cited knowledge, contributor histories, and teams using 47,959 articles from six fields over 2010-2025, 51,736 resolved cited works, and chronologically reconstructed prior publication histories for 1,754 randomly selected index contributors. From 2010 to 2022, team size increased by an estimated 37.3% (95% confidence interval [34.4%, 40.3%]), while paper topic breadth declined by 0.0144 on a 0-1 hierarchical distance scale. Cited knowledge was stable to modestly broader, revealing a divergence between focused outputs and the reach of knowledge inputs. Established contributors' prior breadth increased by 0.0190 [-0.0078, 0.0459] by 2019-2022, within a +/-0.05 equivalence bound assessed in sensitivity analysis. In mature citation windows, one standard deviation of focal depth was associated with 8.2% higher 1 + FWCI [1.9%, 14.9%]; average breadth and interaction associations were smaller under the specified equivalence bounds. Post-2022 deviations from earlier trends were not systematic, and recent changes did not vary clearly with baseline AI intensity across 83 subfields. The findings support a differentiated structure of scientific expertise in which focused individual accumulation coexists with expanding collaboration and sustained access to diverse knowledge inputs.
Xiaoshn Nee, Haobo Zhong, Xiaomin Ni
Independent researcher · HSBC Business School, Peking University, Shenzhen City, Guangdong 518055, China · Artificial Intelligence Research Institute, Shenzhen University of Advanced Technology, Shenzhen, China
The extent to which Artificial Intelligence (AI) technologies can trigger generalized paradigm shifts in science is unclear. Although these technologies have revolutionized data collection and analysis in specific fields, their overall impact depends on the scope and ways of adoption. We analyze over 227 million scholarly works from the OpenAlex collection (1960-2024) spanning four scientific domains and 46 fields. To distinguish the use of AI as research method (AI adoption) from mentioning AI-related terms (AI engagement), we developed a two-step AI-assisted semantic classification pipeline, validated through human coding of 911 abstracts and a robustness check on 348,000 full-text articles (PLOS One). We document differences in the timing and extent of AI adoption across domains, with generalized exponential growth after 2015. The transformative nature of this growth, however, is less apparent. AI-supported research is confined to a few topics with strong ties to Computer Science and conventional statistical frameworks, suggesting limited epistemological transformation. It is also associated with an unwarranted citation premium and substantially higher retraction rates than non-AI-supported. Geographically, while wealthy countries lead in AI publications per capita, global South countries in a belt from Indonesia to Algeria lead in AI adoption relative to their national output, signaling a distinctive resource concentration pattern. The transformative capacity of AI in science thus remains untapped, and its rapid adoption underlines challenges in research openness, transparency, reproducibility, and ethics. We discuss how best research practices could boost the benefits of AI adoption and highlight areas that warrant closer scrutiny.
Andrés F. Castro Torres, Joan Giner-Miguelez, Mercè Crosas
Computational Social Science and Humanities, Barcelona Supercomputing Center, BSC-CNS, Barcelona, Spain · Max Planck Institute for Demographic Research, Rostock, Germany