{"id":"5535360645e227335c4f8df2fed1fc25599998f69e3af74ea8b2da26653c1e76","title":"Arvind Narayanan and Sayash Kapoor argue in AI as Normal Technology that adoption shapes AI’s effects","argument":"Published April 15, 2025, Arvind Narayanan and Sayash Kapoor present “AI as normal technology” as a description of AI, a forecast and a policy stance. They argue that advances in AI methods do not by themselves transform work or society: applications must be developed and adopted, a process they expect to unfold gradually. Benchmarks, they contend, can overstate real-world effects when tests miss the complexity and judgment involved in professional work.\n\nThe authors contrast AI as a tool people can control with visions of AI as a separate, potentially superintelligent entity. They predict that more jobs will involve controlling AI, while warning that handing systems larger tasks “end-to-end” can reduce legibility and control. If AI follows the normal-technology path, they argue, systemic risks such as inequality and concentrated power deserve more attention than catastrophic risks they consider speculative.\n\nTheir policy conclusion is to prioritize resilience and reduce uncertainty. They argue that nonproliferation could concentrate power and create single points of failure. Policy, they say, can hinder or support AI diffusion, and should help make its benefits more equitable. They acknowledge their predictions are uncertain and unquantified; whether AI’s adoption and impacts follow the gradual pattern they describe remains unknown.","topicIds":["artificial-general-intelligence"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 24, 2026; originally published April 15, 2025. Narayanan and Kapoor say they have not quantified probabilities for their predictions and describe the essay as an initial articulation of their worldview, not a point-by-point response to counterarguments.","speaker":{"name":"Arvind Narayanan and Sayash Kapoor","roleAtSource":"Publisher"},"source":{"title":"AI as Normal Technology","publisher":"Arvind Narayanan and Sayash Kapoor","format":"Article","url":"https://www.normaltech.ai/p/ai-as-normal-technology","publishedAt":"2025-04-15T14:53:59.000Z","dateBasis":"Date supplied by the publisher’s feed."},"evidence":[{"text":"Narayanan and Kapoor frame “AI as normal technology” as both a description of present AI and a forecast and policy stance; they reject treating AI as a separate, humanlike agent.","excerpt":"The statement “AI is normal technology” is three things: a description of current AI, a prediction about the foreseeable future of AI, and a prescription about how we should treat it.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"},{"text":"They argue that AI’s social and economic effects depend on applications and adoption as well as on advances in AI methods, and that these stages unfold at different timescales.","excerpt":"transformative economic and societal impacts will be slow (on the timescale of decades), making a critical distinction between AI methods, AI applications, and AI adoption, arguing that the three happen at different timescales.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"},{"text":"They argue that benchmark results can overstate real-world impact because tests may not measure the complex work and judgment involved in professional practice.","excerpt":"The easier a task is to measure via benchmarks, the less likely it is to represent the kind of complex, contextual work that defines professional practice.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"},{"text":"They predict that a growing share of jobs and tasks will involve controlling AI, while warning that more autonomous, end-to-end systems can reduce legibility and control.","excerpt":"As more physical and cognitive tasks become amenable to automation, we predict that an increasing percentage of human jobs and tasks will be related to AI control.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"},{"text":"They argue that, under the normal-technology view, systemic risks such as inequality and concentration of power matter more than catastrophic risks they consider speculative.","excerpt":"If AI is normal technology, these risks become far more important than the catastrophic ones discussed above.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"},{"text":"They recommend resilience as the policy strategy and argue that nonproliferation may increase concentration and brittleness.","excerpt":"the strategy that policymakers should center is resilience, which consists of taking actions now to improve our ability to deal with unexpected developments in the future.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.normaltech.ai/p/ai-as-normal-technology","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"What evidence would show whether AI’s effects are unfolding gradually through application development and adoption, rather than through sudden economic change?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Arvind Narayanan and Sayash Kapoor argue in AI as Normal Technology that adoption shapes AI’s effects","summary":"Published April 15, 2025, Arvind Narayanan and Sayash Kapoor present “AI as normal technology” as a description of AI, a forecast and a policy stance. They argue that advances in AI methods do not by themselves transform work or society: applications must be developed and adopted, a process they expect to unfold gradually. Benchmarks, they contend, can overstate real-world effects when tests miss the complexity and judgment involved in professional work.\n\nThe authors contrast AI as a tool people can control with visions of AI as a separate, potentially superintelligent entity. They predict that more jobs will involve controlling AI, while warning that handing systems larger tasks “end-to-end” can reduce legibility and control. If AI follows the normal-technology path, they argue, systemic risks such as inequality and concentrated power deserve more attention than catastrophic risks they consider speculative.\n\nTheir policy conclusion is to prioritize resilience and reduce uncertainty. They argue that nonproliferation could concentrate power and create single points of failure. Policy, they say, can hinder or support AI diffusion, and should help make its benefits more equitable. They acknowledge their predictions are uncertain and unquantified; whether AI’s adoption and impacts follow the gradual pattern they describe remains unknown.","question":"What evidence would show whether AI’s effects are unfolding gradually through application development and adoption, rather than through sudden economic change?","claims":[{"text":"Narayanan and Kapoor frame “AI as normal technology” as both a description of present AI and a forecast and policy stance; they reject treating AI as a separate, humanlike agent.","excerpt":"The statement “AI is normal technology” is three things: a description of current AI, a prediction about the foreseeable future of AI, and a prescription about how we should treat it.","support":"source-stated"},{"text":"They argue that AI’s social and economic effects depend on applications and adoption as well as on advances in AI methods, and that these stages unfold at different timescales.","excerpt":"transformative economic and societal impacts will be slow (on the timescale of decades), making a critical distinction between AI methods, AI applications, and AI adoption, arguing that the three happen at different timescales.","support":"source-stated"},{"text":"They argue that benchmark results can overstate real-world impact because tests may not measure the complex work and judgment involved in professional practice.","excerpt":"The easier a task is to measure via benchmarks, the less likely it is to represent the kind of complex, contextual work that defines professional practice.","support":"source-stated"},{"text":"They predict that a growing share of jobs and tasks will involve controlling AI, while warning that more autonomous, end-to-end systems can reduce legibility and control.","excerpt":"As more physical and cognitive tasks become amenable to automation, we predict that an increasing percentage of human jobs and tasks will be related to AI control.","support":"source-stated"},{"text":"They argue that, under the normal-technology view, systemic risks such as inequality and concentration of power matter more than catastrophic risks they consider speculative.","excerpt":"If AI is normal technology, these risks become far more important than the catastrophic ones discussed above.","support":"source-stated"},{"text":"They recommend resilience as the policy strategy and argue that nonproliferation may increase concentration and brittleness.","excerpt":"the strategy that policymakers should center is resilience, which consists of taking actions now to improve our ability to deal with unexpected developments in the future.","support":"source-stated"}],"limitations":["Text as retrieved on October 24, 2026; originally published April 15, 2025.","Narayanan and Kapoor say they have not quantified probabilities for their predictions and describe the essay as an initial articulation of their worldview, not a point-by-point response to counterarguments."],"mediaId":null,"correctionReason":null,"id":"5535360645e227335c4f8df2fed1fc25599998f69e3af74ea8b2da26653c1e76","publicationId":"6193e856-64bd-c6d2-b68a-729d0a399614","version":1,"sourceUrl":"https://www.normaltech.ai/p/ai-as-normal-technology","sourceTitle":"AI as Normal Technology","publisher":"Arvind Narayanan and Sayash Kapoor","publishedAt":"2025-04-15T14:53:59.000Z","originalPublished":{"value":"2025-04-15T14:53:59.000Z","precision":"instant","basis":"backfill:feed_published"},"firstObservedAt":"2026-10-24T00:48:07.187Z","createdAt":"2026-11-14T18:43:24.368Z","historical":true,"entityIds":[],"topicIds":["artificial-general-intelligence"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.normaltech.ai/p/ai-as-normal-technology","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"8a9a66369c1852590dc2b48eb57812e136e422b5eb25b50b689daaef0e985cc8","rubricVersion":"unrecorded","provider":"unrecorded","model":"anthropic/claude-opus-5.5","reasonCodes":["bridge_admitted_by_model","second_opinion_categorical"],"confidence":1,"decidedAt":"2026-10-24T00:48:03.574Z"},"provenance":{"textHash":"6f73f809d68628b0e49910fe9d209306aaaa41a61a87e32b39ef6455ed069922","observationId":"96f559cc2409db93a9aa1f2c67dbc0762ad1130be4d5d71e46e9bbcc816df8f3","retrievedAt":"2026-10-24T00:49:35.275Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"0a017194-a7d6-4e19-daea-71aa8087b7eb"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"d0aba929-1bd1-cd56-82fd-dce57dcfd83d"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-14T18:38:49.411Z","model":"anthropic/claude-opus-5.5"}},"canonicalUrl":"https://escapemeridian.com/readings/arvind-narayanan-and-sayash-kapoor-argue-in-ai-as-normal-55353606","formats":{"html":"/readings/arvind-narayanan-and-sayash-kapoor-argue-in-ai-as-normal-55353606","markdown":"/readings/arvind-narayanan-and-sayash-kapoor-argue-in-ai-as-normal-55353606.md","json":"/readings/arvind-narayanan-and-sayash-kapoor-argue-in-ai-as-normal-55353606.json"},"attribution":"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.","maintainedGuideLinks":[],"conditions":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-benchmark","name":"Benchmark","explanation":"A defined set of tests used to compare systems. Results depend on the tasks and measures selected.","referenceUrl":"https://arxiv.org/abs/2211.09110"},{"id":"term-superintelligence","name":"Superintelligence","explanation":"A hypothetical AI far more capable than people across most intellectual tasks.","referenceUrl":"https://en.wikipedia.org/wiki/Superintelligence"},{"id":"term-systemic-risk","name":"Systemic risk","explanation":"A risk that can spread through a whole system, such as a financial market, rather than staying in one place.","referenceUrl":"https://en.wikipedia.org/wiki/Systemic_risk"}],"entities":[]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"5535360645e227335c4f8df2fed1fc25599998f69e3af74ea8b2da26653c1e76","revision":"6193e856-64bd-c6d2-b68a-729d0a399614","revisionKind":"publication-id","title":"Arvind Narayanan and Sayash Kapoor argue in AI as Normal Technology that adoption shapes AI’s effects","url":"https://escapemeridian.com/readings/arvind-narayanan-and-sayash-kapoor-argue-in-ai-as-normal-55353606"},"source":{"type":"source","id":"https://www.normaltech.ai/p/ai-as-normal-technology","revision":"6f73f809d68628b0e49910fe9d209306aaaa41a61a87e32b39ef6455ed069922","revisionKind":"sha256-source-text","title":"AI as Normal Technology","url":"https://www.normaltech.ai/p/ai-as-normal-technology"},"status":"available","items":[],"next":null,"scope":"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."},"revisions":[{"id":"5535360645e227335c4f8df2fed1fc25599998f69e3af74ea8b2da26653c1e76","title":"Arvind Narayanan and Sayash Kapoor argue in AI as Normal Technology that adoption shapes AI’s effects","argument":"Published April 15, 2025, Arvind Narayanan and Sayash Kapoor present “AI as normal technology” as a description of AI, a forecast and a policy stance. They argue that advances in AI methods do not by themselves transform work or society: applications must be developed and adopted, a process they expect to unfold gradually. Benchmarks, they contend, can overstate real-world effects when tests miss the complexity and judgment involved in professional work.\n\nThe authors contrast AI as a tool people can control with visions of AI as a separate, potentially superintelligent entity. They predict that more jobs will involve controlling AI, while warning that handing systems larger tasks “end-to-end” can reduce legibility and control. If AI follows the normal-technology path, they argue, systemic risks such as inequality and concentrated power deserve more attention than catastrophic risks they consider speculative.\n\nTheir policy conclusion is to prioritize resilience and reduce uncertainty. They argue that nonproliferation could concentrate power and create single points of failure. Policy, they say, can hinder or support AI diffusion, and should help make its benefits more equitable. They acknowledge their predictions are uncertain and unquantified; whether AI’s adoption and impacts follow the gradual pattern they describe remains unknown.","topicIds":["artificial-general-intelligence"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 24, 2026; originally published April 15, 2025. Narayanan and Kapoor say they have not quantified probabilities for their predictions and describe the essay as an initial articulation of their worldview, not a point-by-point response to counterarguments.","speaker":{"name":"Arvind Narayanan and Sayash Kapoor","roleAtSource":"Publisher"},"source":{"title":"AI as Normal Technology","publisher":"Arvind Narayanan and Sayash Kapoor","format":"Article","url":"https://www.normaltech.ai/p/ai-as-normal-technology","publishedAt":"2025-04-15T14:53:59.000Z","dateBasis":"Date supplied by the publisher’s feed."},"evidence":[{"text":"Narayanan and Kapoor frame “AI as normal technology” as both a description of present AI and a forecast and policy stance; they reject treating AI as a separate, humanlike agent.","excerpt":"The statement “AI is normal technology” is three things: a description of current AI, a prediction about the foreseeable future of AI, and a prescription about how we should treat it.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"},{"text":"They argue that AI’s social and economic effects depend on applications and adoption as well as on advances in AI methods, and that these stages unfold at different timescales.","excerpt":"transformative economic and societal impacts will be slow (on the timescale of decades), making a critical distinction between AI methods, AI applications, and AI adoption, arguing that the three happen at different timescales.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"},{"text":"They argue that benchmark results can overstate real-world impact because tests may not measure the complex work and judgment involved in professional practice.","excerpt":"The easier a task is to measure via benchmarks, the less likely it is to represent the kind of complex, contextual work that defines professional practice.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"},{"text":"They predict that a growing share of jobs and tasks will involve controlling AI, while warning that more autonomous, end-to-end systems can reduce legibility and control.","excerpt":"As more physical and cognitive tasks become amenable to automation, we predict that an increasing percentage of human jobs and tasks will be related to AI control.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"},{"text":"They argue that, under the normal-technology view, systemic risks such as inequality and concentration of power matter more than catastrophic risks they consider speculative.","excerpt":"If AI is normal technology, these risks become far more important than the catastrophic ones discussed above.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"},{"text":"They recommend resilience as the policy strategy and argue that nonproliferation may increase concentration and brittleness.","excerpt":"the strategy that policymakers should center is resilience, which consists of taking actions now to improve our ability to deal with unexpected developments in the future.","url":"https://www.normaltech.ai/p/ai-as-normal-technology","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.normaltech.ai/p/ai-as-normal-technology","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"What evidence would show whether AI’s effects are unfolding gradually through application development and adoption, rather than through sudden economic change?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Arvind Narayanan and Sayash Kapoor argue in AI as Normal Technology that adoption shapes AI’s effects","summary":"Published April 15, 2025, Arvind Narayanan and Sayash Kapoor present “AI as normal technology” as a description of AI, a forecast and a policy stance. They argue that advances in AI methods do not by themselves transform work or society: applications must be developed and adopted, a process they expect to unfold gradually. Benchmarks, they contend, can overstate real-world effects when tests miss the complexity and judgment involved in professional work.\n\nThe authors contrast AI as a tool people can control with visions of AI as a separate, potentially superintelligent entity. They predict that more jobs will involve controlling AI, while warning that handing systems larger tasks “end-to-end” can reduce legibility and control. If AI follows the normal-technology path, they argue, systemic risks such as inequality and concentrated power deserve more attention than catastrophic risks they consider speculative.\n\nTheir policy conclusion is to prioritize resilience and reduce uncertainty. They argue that nonproliferation could concentrate power and create single points of failure. Policy, they say, can hinder or support AI diffusion, and should help make its benefits more equitable. They acknowledge their predictions are uncertain and unquantified; whether AI’s adoption and impacts follow the gradual pattern they describe remains unknown.","question":"What evidence would show whether AI’s effects are unfolding gradually through application development and adoption, rather than through sudden economic change?","claims":[{"text":"Narayanan and Kapoor frame “AI as normal technology” as both a description of present AI and a forecast and policy stance; they reject treating AI as a separate, humanlike agent.","excerpt":"The statement “AI is normal technology” is three things: a description of current AI, a prediction about the foreseeable future of AI, and a prescription about how we should treat it.","support":"source-stated"},{"text":"They argue that AI’s social and economic effects depend on applications and adoption as well as on advances in AI methods, and that these stages unfold at different timescales.","excerpt":"transformative economic and societal impacts will be slow (on the timescale of decades), making a critical distinction between AI methods, AI applications, and AI adoption, arguing that the three happen at different timescales.","support":"source-stated"},{"text":"They argue that benchmark results can overstate real-world impact because tests may not measure the complex work and judgment involved in professional practice.","excerpt":"The easier a task is to measure via benchmarks, the less likely it is to represent the kind of complex, contextual work that defines professional practice.","support":"source-stated"},{"text":"They predict that a growing share of jobs and tasks will involve controlling AI, while warning that more autonomous, end-to-end systems can reduce legibility and control.","excerpt":"As more physical and cognitive tasks become amenable to automation, we predict that an increasing percentage of human jobs and tasks will be related to AI control.","support":"source-stated"},{"text":"They argue that, under the normal-technology view, systemic risks such as inequality and concentration of power matter more than catastrophic risks they consider speculative.","excerpt":"If AI is normal technology, these risks become far more important than the catastrophic ones discussed above.","support":"source-stated"},{"text":"They recommend resilience as the policy strategy and argue that nonproliferation may increase concentration and brittleness.","excerpt":"the strategy that policymakers should center is resilience, which consists of taking actions now to improve our ability to deal with unexpected developments in the future.","support":"source-stated"}],"limitations":["Text as retrieved on October 24, 2026; originally published April 15, 2025.","Narayanan and Kapoor say they have not quantified probabilities for their predictions and describe the essay as an initial articulation of their worldview, not a point-by-point response to counterarguments."],"mediaId":null,"correctionReason":null,"id":"5535360645e227335c4f8df2fed1fc25599998f69e3af74ea8b2da26653c1e76","publicationId":"6193e856-64bd-c6d2-b68a-729d0a399614","version":1,"sourceUrl":"https://www.normaltech.ai/p/ai-as-normal-technology","sourceTitle":"AI as Normal Technology","publisher":"Arvind Narayanan and Sayash Kapoor","publishedAt":"2025-04-15T14:53:59.000Z","originalPublished":{"value":"2025-04-15T14:53:59.000Z","precision":"instant","basis":"backfill:feed_published"},"firstObservedAt":"2026-10-24T00:48:07.187Z","createdAt":"2026-11-14T18:43:24.368Z","historical":true,"entityIds":[],"topicIds":["artificial-general-intelligence"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.normaltech.ai/p/ai-as-normal-technology","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"8a9a66369c1852590dc2b48eb57812e136e422b5eb25b50b689daaef0e985cc8","rubricVersion":"unrecorded","provider":"unrecorded","model":"anthropic/claude-opus-5.5","reasonCodes":["bridge_admitted_by_model","second_opinion_categorical"],"confidence":1,"decidedAt":"2026-10-24T00:48:03.574Z"},"provenance":{"textHash":"6f73f809d68628b0e49910fe9d209306aaaa41a61a87e32b39ef6455ed069922","observationId":"96f559cc2409db93a9aa1f2c67dbc0762ad1130be4d5d71e46e9bbcc816df8f3","retrievedAt":"2026-10-24T00:49:35.275Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"0a017194-a7d6-4e19-daea-71aa8087b7eb"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"d0aba929-1bd1-cd56-82fd-dce57dcfd83d"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-14T18:38:49.411Z","model":"anthropic/claude-opus-5.5"}}}],"audits":[{"publicationId":"6193e856-64bd-c6d2-b68a-729d0a399614","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading consistently attributes the paper's arguments, predictions and recommendations to Arvind Narayanan and Sayash Kapoor, matching the byline. It presents them as the authors' claims, not as established fact.","excerpt":"We articulate a vision of artificial intelligence (AI) as normal technology."},{"dimension":"summary","status":"uncertain","explanation":"The summary is faithful overall and keeps the authors' uncertainty. However, the policy paragraph leans on 'nonproliferation' without saying what it means here: limiting who can obtain powerful AI, for example through licensing, export controls or bans on open-weight models. 'Diffusion' is also unexplained. Neither term is in the definitions list, so a general reader cannot follow the central policy argument. Smallest repair: add one plain-language clause defining nonproliferation and a tooltip for diffusion.","excerpt":"Nonproliferation policies seek to limit the number of actors who can obtain powerful AI capabilities."},{"dimension":"date","status":"pass","explanation":"The reading states April 15, 2025, which matches originalPublished and the dateline at day precision. The page contains a later 'Update (September 2025)' note, but the reading neither uses nor relies on it. Forecasts are framed as the authors' predictions, and there is no hindsight framing. This is a publisher page, so no archive disclosure is required.","excerpt":"Arvind Narayanan and Sayash Kapoor\nApr 15, 2025"},{"dimension":"support","status":"pass","explanation":"Each claim is source-stated and backed by its excerpt. The end-to-end warning, equitable benefits and the nonproliferation critique all appear in the text. One small imprecision: 'catastrophic risks they consider speculative' should be read restrictively. The source calls catastrophic misalignment the most speculative risk, not every catastrophic risk. This is not a contradiction.","excerpt":"In the view of AI as normal technology, catastrophic misalignment is (by far) the most speculative of the risks that we discuss."},{"dimension":"media","status":"pass","explanation":"No media is selected and no verified candidates were supplied. A null media choice is appropriate.","excerpt":null}],"createdAt":"2026-11-14T18:38:49.411Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}