Beichen Lu, Honglin Bao, Qi Huang, Qingguo Meng, James A. Evans · 2026 · arXiv (Cornell University)
DOI: 10.48550/arxiv.2610.27296 · cited by 0 works (OpenAlex) · open-access PDF
El progreso en la inteligencia artificial se asume ampliamente para escalar con el esfuerzo: más investigadores, papeles e inversiones deben acelerar el descubrimiento. Las teorías del crecimiento tecnológico en cambio sitúan la dificultad del descubrimiento en el espacio de las ideas. Aquí mostramos que ni la escala ni el espacio son suficientes: los avances dependen de la atención diversa, que la concentración colectiva puede agotar. A través de 36 años de ciencia de la computación, analizamos 7.2 millones de artículos, 73.089 revisiones de pares y 3.7 millones de investigadores y sus cambios de tema. El campo se expandió mientras su atención se restringió de forma aguda después del Transformer. Más de la mitad de los investigadores que cambiaron de tema se trasladaron a la IA, especialmente a los de mediana carrera, y la confianza
Machine translation (Meta M2M100); the English original follows.
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.
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
Topics: Field (mathematics), Diversity (politics), Novelty, Scale (ratio), Space (punctuation), Point (geometry)
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.27296, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange