# Sayash Kapoor and Arvind Narayanan argue in Normal Tech that AI may slow science [Canonical reading](https://escapemeridian.com/readings/sayash-kapoor-and-arvind-narayanan-argue-in-normal-tech-that-ai-c8707c1c) On July 16, 2025, Sayash Kapoor and Arvind Narayanan argue that AI could slow scientific progress even when it helps individual scientists. They call the mismatch between growing research output and slower progress the production-progress paradox. Their concern is that institutions reward measurable production, such as papers, while breakthroughs and understanding are harder to measure. If AI makes research production easier without addressing those incentives, they argue, it could worsen the gap and make it harder for novel ideas to gain attention. They also warn that AI-based modeling can improve prediction without improving understanding, and that errors in software-based research can waste researchers’ time. The authors say the outcome may differ across fields and is “certainly not a foregone conclusion.” They argue that journals, universities, funders and policymakers have more power than individual researchers to change incentives and quality control. They call for evaluating AI tools by their effects on understanding and scientific communities, not only time saved and output quality. They close with optimism that scientific norms and processes will catch up, but warn that “for now, it’s going to be a bumpy ride.” How much the slowdown reflects production-focused incentives, rather than other causes, remains unclear. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source Arvind Narayanan and Sayash Kapoor · Article Original source date: Jul 16, 2025 (2025-07-16T21:35:56.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-11-13T02:13:23.605Z. First observed: 2026-10-24T00:56:06.154Z. Retrieved: 2026-10-24T00:57:37.590Z. - [Could AI slow science?](https://www.normaltech.ai/p/could-ai-slow-science) ## Limits Text as retrieved on October 24, 2026; originally published July 16, 2025. The authors say progress has no objective metric, and the measures they discuss have shortcomings. They also say it is less clear what is happening to progress at an aggregate level and what drives the slowdown. The essay argues that AI may worsen the production-progress paradox, but says this may not be true in all scientific fields and is not a foregone conclusion. ## Open question What changes to scientific incentives or institutions would improve progress without replacing one flawed measure with another? ## Source-stated Sayash Kapoor and Arvind Narayanan argue that AI could worsen the gap between rising scientific output and slower progress, while noting that the outcome may differ across fields and is not inevitable. > Our analysis in this essay suggests that AI is likely to worsen the gap. This may not be true in all scientific fields, and it is certainly not a foregone conclusion. - [Supporting source](https://www.normaltech.ai/p/could-ai-slow-science) ## Source-stated The authors argue that production-focused incentives may discourage risky, novel work by rewarding researchers for measurable outputs such as papers and grant funding. > universities and other scientific institutions judge researchers based on measurable criteria such as how many papers they publish and the amount of grant funding they receive. - [Supporting source](https://www.normaltech.ai/p/could-ai-slow-science) ## Source-stated The authors warn that AI-based modeling may improve prediction without advancing understanding, potentially keeping flawed theories in use longer. > if AI allows us to make better predictions from incorrect theories, it might slow down scientific progress if this results in researchers using flawed theories for longer. - [Supporting source](https://www.normaltech.ai/p/could-ai-slow-science) ## Source-stated The authors say scientific progress is difficult to measure and that the available measures of progress have shortcomings. > since progress doesn't have an objective metric, we need to rely on proxies for measuring it, and these proxies will inevitably have some flaws. - [Supporting source](https://www.normaltech.ai/p/could-ai-slow-science) ## Source-stated Sayash Kapoor and Arvind Narayanan argue that scientific norms and processes may catch up over time, but describe the near-term as difficult. > We are optimistic that scientific norms and processes will catch up in the long run. But for now, it’s going to be a bumpy ride. - [Supporting source](https://www.normaltech.ai/p/could-ai-slow-science) ## Source and contributor context Arvind Narayanan and Sayash Kapoor · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## TIME TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services. - [Explore this context](/entities/time) - [Reference](https://time.com/) ## Editorial provenance Version 1. Publication ID: cf68dcad-c73e-27d6-b381-459399f8c6d5. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-13T02:08:38.423Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: 4e7aa08c195a1341ad785caefb76bce83aba402d83f937fa8b4d94a220dab005. Source text hash: 1b0e4f8ee6d79cca5c6c4f75be203595a9db9d5889023ce36c5deadc5c617e20. ## Revision and source-check history ### Version 1 — 2026-11-13T02:13:23.605Z On July 16, 2025, Sayash Kapoor and Arvind Narayanan argue that AI could slow scientific progress even when it helps individual scientists. They call the mismatch between growing research output and slower progress the production-progress paradox. Their concern is that institutions reward measurable production, such as papers, while breakthroughs and understanding are harder to measure. If AI makes research production easier without addressing those incentives, they argue, it could worsen the gap and make it harder for novel ideas to gain attention. They also warn that AI-based modeling can improve prediction without improving understanding, and that errors in software-based research can waste researchers’ time. The authors say the outcome may differ across fields and is “certainly not a foregone conclusion.” They argue that journals, universities, funders and policymakers have more power than individual researchers to change incentives and quality control. They call for evaluating AI tools by their effects on understanding and scientific communities, not only time saved and output quality. They close with optimism that scientific norms and processes will catch up, but warn that “for now, it’s going to be a bumpy ride.” How much the slowdown reflects production-focused incentives, rather than other causes, remains unclear. ## Original source Arvind Narayanan and Sayash Kapoor · Article Original source date: Jul 16, 2025 (2025-07-16T21:35:56.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-11-13T02:13:23.605Z. First observed: 2026-10-24T00:56:06.154Z. Retrieved: 2026-10-24T00:57:37.590Z. - [Could AI slow science?](https://www.normaltech.ai/p/could-ai-slow-science) ## Limits Text as retrieved on October 24, 2026; originally published July 16, 2025. The authors say progress has no objective metric, and the measures they discuss have shortcomings. They also say it is less clear what is happening to progress at an aggregate level and what drives the slowdown. The essay argues that AI may worsen the production-progress paradox, but says this may not be true in all scientific fields and is not a foregone conclusion. ## Open question What changes to scientific incentives or institutions would improve progress without replacing one flawed measure with another? ## Source-stated Sayash Kapoor and Arvind Narayanan argue that AI could worsen the gap between rising scientific output and slower progress, while noting that the outcome may differ across fields and is not inevitable. > Our analysis in this essay suggests that AI is likely to worsen the gap. This may not be true in all scientific fields, and it is certainly not a foregone conclusion. - [Supporting source](https://www.normaltech.ai/p/could-ai-slow-science) ## Source-stated The authors argue that production-focused incentives may discourage risky, novel work by rewarding researchers for measurable outputs such as papers and grant funding. > universities and other scientific institutions judge researchers based on measurable criteria such as how many papers they publish and the amount of grant funding they receive. - [Supporting source](https://www.normaltech.ai/p/could-ai-slow-science) ## Source-stated The authors warn that AI-based modeling may improve prediction without advancing understanding, potentially keeping flawed theories in use longer. > if AI allows us to make better predictions from incorrect theories, it might slow down scientific progress if this results in researchers using flawed theories for longer. - [Supporting source](https://www.normaltech.ai/p/could-ai-slow-science) ## Source-stated The authors say scientific progress is difficult to measure and that the available measures of progress have shortcomings. > since progress doesn't have an objective metric, we need to rely on proxies for measuring it, and these proxies will inevitably have some flaws. - [Supporting source](https://www.normaltech.ai/p/could-ai-slow-science) ## Source-stated Sayash Kapoor and Arvind Narayanan argue that scientific norms and processes may catch up over time, but describe the near-term as difficult. > We are optimistic that scientific norms and processes will catch up in the long run. But for now, it’s going to be a bumpy ride. - [Supporting source](https://www.normaltech.ai/p/could-ai-slow-science) ## Source and contributor context Arvind Narayanan and Sayash Kapoor · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## TIME TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services. - [Explore this context](/entities/time) - [Reference](https://time.com/) ## Editorial provenance Version 1. Publication ID: cf68dcad-c73e-27d6-b381-459399f8c6d5. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-13T02:08:38.423Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: 4e7aa08c195a1341ad785caefb76bce83aba402d83f937fa8b4d94a220dab005. Source text hash: 1b0e4f8ee6d79cca5c6c4f75be203595a9db9d5889023ce36c5deadc5c617e20. ### passed — 2026-11-13T02:08:38.423Z — anthropic/claude-opus-5.5 Publication: cf68dcad-c73e-27d6-b381-459399f8c6d5 attribution — pass: The essay's byline names Sayash Kapoor and Arvind Narayanan. The reading attributes its arguments to them throughout, and the normaltech.ai URL is consistent with the "Normal Tech" label in the title. > Sayash Kapoor and Arvind Narayanan Jul 16, 2025 summary — pass: The reading leads with the authors' thesis and explains the production-progress paradox in plain terms. It keeps their hedges, the institutional actors they name, the tool-evaluation proposal and the closing optimism with its bumpy-ride warning. The question follows from the source's Goodhart's law discussion and is not leading. > Any serious attempt to forecast the impact of AI on science must confront the production-progress paradox. date — pass: July 16, 2025 matches the dateline and the UTC day of originalPublished. The source's July 17 update note describes minor wording edits, and the reading does not rely on them. The reading uses no hindsight and keeps the authors' forecasts attributed. > Update (July 17, 2025): Minor wording edits for clarity. support — pass: Each claim is source-stated and carries a matching excerpt. The closing point that the drivers of the slowdown remain unclear is grounded in the authors' own statement. Benefits and costs are not overstated. > we’re far from a consensus story about what’s driving the slowdown. media — pass: No media was selected and none was supplied, so nothing needs to be checked for identity or relevance. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/c8707c1cbaa36134a5ab2f030c956cbafe9929b8c014c6643064ef14eb4755e0/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/sayash-kapoor-and-arvind-narayanan-argue-in-normal-tech-that-ai-c8707c1c.json)