{"id":"e1002b5e1b0c917b3a70314258615697c61165f10b3973aa1fa07a112fec109f","title":"Epoch AI proposes a task map for AI research and development","argument":"Epoch AI published this proposal on June 17, 2026. The authors argue that common ways of forecasting AI research automation rely on indirect measures, such as compute or benchmark performance, that may miss important parts of research work. They propose a detailed task map for AI R&D: a way to describe the work involved and estimate how much current AI automates each task. The initial taxonomy groups tasks into categories and gives each an automation rating. The authors say such a map could help researchers interpret benchmark results, identify tasks that are not being measured, and study how AI supports researchers. They stress that the ratings are subjective and that a task list alone cannot show how well AI performs on each task. They also note that AI systems might make progress through workflows unlike current human research, which a human-work-based taxonomy could miss. The proposal is an initial attempt, not a measure of overall research automation. How could the task descriptions and automation ratings be tested and updated as AI research work changes?","topicIds":["artificial-general-intelligence"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 23, 2026; originally published June 17, 2026. The authors describe the taxonomy and automation ratings as an initial, subjective attempt; the evidence does not establish how accurately the ratings measure AI performance. The authors note that a taxonomy based on current human workflows could miss AI progress through different routes.","speaker":{"name":"Epoch AI","roleAtSource":"Publisher"},"source":{"title":"Toward an O*NET for AI R&D","publisher":"Epoch AI","format":"Article","url":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","publishedAt":"2026-06-17T00:00:00.000Z","dateBasis":"Date supplied by the publisher’s feed."},"evidence":[{"text":"The authors say AI forecasting often extrapolates trends in easily measured proxies rather than directly measuring how close AI is to doing AI research.","excerpt":"the trends they extrapolate share a common weakness: they lean heavily on easy-to-measure things, not what we directly care about — how close AI is to doing AI research itself.","url":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","relationship":"Source-stated"},{"text":"The authors present a first taxonomy of AI R&D work, divided into six categories and over sixty tasks, with ratings of how much current AI automates each task.","excerpt":"six categories spanning a frontier AI company’s research workflow, based on literature review and brainstorming. The categories are broken into over sixty tasks, each rated from 0 to 5 on how much we think current AIs automate it.","url":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","relationship":"Source-stated"},{"text":"The authors say the ratings are subjective and that future evaluation methods, such as internal benchmarks, could improve them.","excerpt":"This is of course quite subjective and there’s a ton of room for people to develop better ways to evaluate this, such as through internal benchmarks.","url":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","relationship":"Source-stated"},{"text":"The authors warn that AI systems could make progress through routes unlike current human AI R&D, which would leave a human-work-based task list incomplete.","excerpt":"It’s possible that future AI systems make progress through routes that just really do not resemble current human AI R&D. If so, a task list based on current human workflows would completely miss the action","url":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How could the task descriptions and automation ratings be tested and updated as AI research work changes?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Epoch AI proposes a task map for AI research and development","summary":"Epoch AI published this proposal on June 17, 2026. The authors argue that common ways of forecasting AI research automation rely on indirect measures, such as compute or benchmark performance, that may miss important parts of research work. They propose a detailed task map for AI R&D: a way to describe the work involved and estimate how much current AI automates each task. The initial taxonomy groups tasks into categories and gives each an automation rating. The authors say such a map could help researchers interpret benchmark results, identify tasks that are not being measured, and study how AI supports researchers. They stress that the ratings are subjective and that a task list alone cannot show how well AI performs on each task. They also note that AI systems might make progress through workflows unlike current human research, which a human-work-based taxonomy could miss. The proposal is an initial attempt, not a measure of overall research automation. How could the task descriptions and automation ratings be tested and updated as AI research work changes?","question":"How could the task descriptions and automation ratings be tested and updated as AI research work changes?","claims":[{"text":"The authors say AI forecasting often extrapolates trends in easily measured proxies rather than directly measuring how close AI is to doing AI research.","excerpt":"the trends they extrapolate share a common weakness: they lean heavily on easy-to-measure things, not what we directly care about — how close AI is to doing AI research itself.","support":"source-stated"},{"text":"The authors present a first taxonomy of AI R&D work, divided into six categories and over sixty tasks, with ratings of how much current AI automates each task.","excerpt":"six categories spanning a frontier AI company’s research workflow, based on literature review and brainstorming. The categories are broken into over sixty tasks, each rated from 0 to 5 on how much we think current AIs automate it.","support":"source-stated"},{"text":"The authors say the ratings are subjective and that future evaluation methods, such as internal benchmarks, could improve them.","excerpt":"This is of course quite subjective and there’s a ton of room for people to develop better ways to evaluate this, such as through internal benchmarks.","support":"source-stated"},{"text":"The authors warn that AI systems could make progress through routes unlike current human AI R&D, which would leave a human-work-based task list incomplete.","excerpt":"It’s possible that future AI systems make progress through routes that just really do not resemble current human AI R&D. If so, a task list based on current human workflows would completely miss the action","support":"source-stated"}],"limitations":["Text as retrieved on October 23, 2026; originally published June 17, 2026.","The authors describe the taxonomy and automation ratings as an initial, subjective attempt; the evidence does not establish how accurately the ratings measure AI performance.","The authors note that a taxonomy based on current human workflows could miss AI progress through different routes."],"mediaId":null,"correctionReason":null,"id":"e1002b5e1b0c917b3a70314258615697c61165f10b3973aa1fa07a112fec109f","publicationId":"0007cbc8-dbe0-855d-729d-48b84aad2700","version":1,"sourceUrl":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","sourceTitle":"Toward an O*NET for AI R&D","publisher":"Epoch AI","publishedAt":"2026-06-17T00:00:00.000Z","originalPublished":{"value":"2026-06-17T00:00:00.000Z","precision":"instant","basis":"backfill:feed_published"},"firstObservedAt":"2026-10-18T03:15:24.309Z","createdAt":"2026-10-30T06:24:09.023Z","historical":true,"entityIds":["epoch-ai"],"topicIds":["artificial-general-intelligence"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"2a54100b68e78c1139ffc70710a6e6a122d380211d6f188bf91298227b23a8c9","rubricVersion":"unrecorded","provider":"vercel","model":"typesafe-ai/jev","reasonCodes":["bridge_admitted_by_model"],"confidence":0.79,"decidedAt":"2026-10-23T23:37:42.660Z"},"provenance":{"textHash":"d3e6a0d5285749f66a0b08cd21a0738082e38c77c4a81d8b07056fcbbd2fa21b","observationId":"3a3da04a08e5de1901821416affdaafe5517c57e498851e243ed8b643278e7d3","retrievedAt":"2026-10-23T23:39:38.715Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"e92fc3c1-af0b-e7e2-255d-25106d7c189d"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"be313416-dbcd-d507-ca28-c6849775ebb7"}},"audit":{"status":"passed","reviewedAt":"2026-10-30T06:24:06.972Z","model":"anthropic/claude-opus-5.5"}},"canonicalUrl":"https://escapemeridian.com/readings/epoch-ai-proposes-a-task-map-for-ai-research-and-development-e1002b5e","formats":{"html":"/readings/epoch-ai-proposes-a-task-map-for-ai-research-and-development-e1002b5e","markdown":"/readings/epoch-ai-proposes-a-task-map-for-ai-research-and-development-e1002b5e.md","json":"/readings/epoch-ai-proposes-a-task-map-for-ai-research-and-development-e1002b5e.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.","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"}],"entities":[{"id":"epoch-ai","name":"Epoch AI","kind":"organization","context":"Epoch AI studies trends in AI capability, training resources and the infrastructure behind them.","referenceUrls":["https://epoch.ai/","https://epoch.ai/data"],"path":"/entities/epoch-ai","json":"/entities/epoch-ai.json","markdown":"/entities/epoch-ai.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"e1002b5e1b0c917b3a70314258615697c61165f10b3973aa1fa07a112fec109f","revision":"0007cbc8-dbe0-855d-729d-48b84aad2700","revisionKind":"publication-id","title":"Epoch AI proposes a task map for AI research and development","url":"https://escapemeridian.com/readings/epoch-ai-proposes-a-task-map-for-ai-research-and-development-e1002b5e"},"source":{"type":"source","id":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","revision":"d3e6a0d5285749f66a0b08cd21a0738082e38c77c4a81d8b07056fcbbd2fa21b","revisionKind":"sha256-source-text","title":"Toward an O*NET for AI R&D","url":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd"},"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":"e1002b5e1b0c917b3a70314258615697c61165f10b3973aa1fa07a112fec109f","title":"Epoch AI proposes a task map for AI research and development","argument":"Epoch AI published this proposal on June 17, 2026. The authors argue that common ways of forecasting AI research automation rely on indirect measures, such as compute or benchmark performance, that may miss important parts of research work. They propose a detailed task map for AI R&D: a way to describe the work involved and estimate how much current AI automates each task. The initial taxonomy groups tasks into categories and gives each an automation rating. The authors say such a map could help researchers interpret benchmark results, identify tasks that are not being measured, and study how AI supports researchers. They stress that the ratings are subjective and that a task list alone cannot show how well AI performs on each task. They also note that AI systems might make progress through workflows unlike current human research, which a human-work-based taxonomy could miss. The proposal is an initial attempt, not a measure of overall research automation. How could the task descriptions and automation ratings be tested and updated as AI research work changes?","topicIds":["artificial-general-intelligence"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 23, 2026; originally published June 17, 2026. The authors describe the taxonomy and automation ratings as an initial, subjective attempt; the evidence does not establish how accurately the ratings measure AI performance. The authors note that a taxonomy based on current human workflows could miss AI progress through different routes.","speaker":{"name":"Epoch AI","roleAtSource":"Publisher"},"source":{"title":"Toward an O*NET for AI R&D","publisher":"Epoch AI","format":"Article","url":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","publishedAt":"2026-06-17T00:00:00.000Z","dateBasis":"Date supplied by the publisher’s feed."},"evidence":[{"text":"The authors say AI forecasting often extrapolates trends in easily measured proxies rather than directly measuring how close AI is to doing AI research.","excerpt":"the trends they extrapolate share a common weakness: they lean heavily on easy-to-measure things, not what we directly care about — how close AI is to doing AI research itself.","url":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","relationship":"Source-stated"},{"text":"The authors present a first taxonomy of AI R&D work, divided into six categories and over sixty tasks, with ratings of how much current AI automates each task.","excerpt":"six categories spanning a frontier AI company’s research workflow, based on literature review and brainstorming. The categories are broken into over sixty tasks, each rated from 0 to 5 on how much we think current AIs automate it.","url":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","relationship":"Source-stated"},{"text":"The authors say the ratings are subjective and that future evaluation methods, such as internal benchmarks, could improve them.","excerpt":"This is of course quite subjective and there’s a ton of room for people to develop better ways to evaluate this, such as through internal benchmarks.","url":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","relationship":"Source-stated"},{"text":"The authors warn that AI systems could make progress through routes unlike current human AI R&D, which would leave a human-work-based task list incomplete.","excerpt":"It’s possible that future AI systems make progress through routes that just really do not resemble current human AI R&D. If so, a task list based on current human workflows would completely miss the action","url":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How could the task descriptions and automation ratings be tested and updated as AI research work changes?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Epoch AI proposes a task map for AI research and development","summary":"Epoch AI published this proposal on June 17, 2026. The authors argue that common ways of forecasting AI research automation rely on indirect measures, such as compute or benchmark performance, that may miss important parts of research work. They propose a detailed task map for AI R&D: a way to describe the work involved and estimate how much current AI automates each task. The initial taxonomy groups tasks into categories and gives each an automation rating. The authors say such a map could help researchers interpret benchmark results, identify tasks that are not being measured, and study how AI supports researchers. They stress that the ratings are subjective and that a task list alone cannot show how well AI performs on each task. They also note that AI systems might make progress through workflows unlike current human research, which a human-work-based taxonomy could miss. The proposal is an initial attempt, not a measure of overall research automation. How could the task descriptions and automation ratings be tested and updated as AI research work changes?","question":"How could the task descriptions and automation ratings be tested and updated as AI research work changes?","claims":[{"text":"The authors say AI forecasting often extrapolates trends in easily measured proxies rather than directly measuring how close AI is to doing AI research.","excerpt":"the trends they extrapolate share a common weakness: they lean heavily on easy-to-measure things, not what we directly care about — how close AI is to doing AI research itself.","support":"source-stated"},{"text":"The authors present a first taxonomy of AI R&D work, divided into six categories and over sixty tasks, with ratings of how much current AI automates each task.","excerpt":"six categories spanning a frontier AI company’s research workflow, based on literature review and brainstorming. The categories are broken into over sixty tasks, each rated from 0 to 5 on how much we think current AIs automate it.","support":"source-stated"},{"text":"The authors say the ratings are subjective and that future evaluation methods, such as internal benchmarks, could improve them.","excerpt":"This is of course quite subjective and there’s a ton of room for people to develop better ways to evaluate this, such as through internal benchmarks.","support":"source-stated"},{"text":"The authors warn that AI systems could make progress through routes unlike current human AI R&D, which would leave a human-work-based task list incomplete.","excerpt":"It’s possible that future AI systems make progress through routes that just really do not resemble current human AI R&D. If so, a task list based on current human workflows would completely miss the action","support":"source-stated"}],"limitations":["Text as retrieved on October 23, 2026; originally published June 17, 2026.","The authors describe the taxonomy and automation ratings as an initial, subjective attempt; the evidence does not establish how accurately the ratings measure AI performance.","The authors note that a taxonomy based on current human workflows could miss AI progress through different routes."],"mediaId":null,"correctionReason":null,"id":"e1002b5e1b0c917b3a70314258615697c61165f10b3973aa1fa07a112fec109f","publicationId":"0007cbc8-dbe0-855d-729d-48b84aad2700","version":1,"sourceUrl":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","sourceTitle":"Toward an O*NET for AI R&D","publisher":"Epoch AI","publishedAt":"2026-06-17T00:00:00.000Z","originalPublished":{"value":"2026-06-17T00:00:00.000Z","precision":"instant","basis":"backfill:feed_published"},"firstObservedAt":"2026-10-18T03:15:24.309Z","createdAt":"2026-10-30T06:24:09.023Z","historical":true,"entityIds":["epoch-ai"],"topicIds":["artificial-general-intelligence"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://epoch.ai/gradient-updates/toward-an-onet-for-ai-rnd","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"2a54100b68e78c1139ffc70710a6e6a122d380211d6f188bf91298227b23a8c9","rubricVersion":"unrecorded","provider":"vercel","model":"typesafe-ai/jev","reasonCodes":["bridge_admitted_by_model"],"confidence":0.79,"decidedAt":"2026-10-23T23:37:42.660Z"},"provenance":{"textHash":"d3e6a0d5285749f66a0b08cd21a0738082e38c77c4a81d8b07056fcbbd2fa21b","observationId":"3a3da04a08e5de1901821416affdaafe5517c57e498851e243ed8b643278e7d3","retrievedAt":"2026-10-23T23:39:38.715Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"e92fc3c1-af0b-e7e2-255d-25106d7c189d"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"be313416-dbcd-d507-ca28-c6849775ebb7"}},"audit":{"status":"passed","reviewedAt":"2026-10-30T06:24:06.972Z","model":"anthropic/claude-opus-5.5"}}}],"audits":[{"publicationId":"0007cbc8-dbe0-855d-729d-48b84aad2700","verdict":"passed","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading correctly attributes the proposal, its arguments and its caveats to Epoch AI and its authors, consistent with the publisher page and first-person authorial text.","excerpt":"In this post, we present a first version of such a taxonomy with six categories spanning a frontier AI company’s research workflow"},{"dimension":"summary","status":"pass","explanation":"The reading leads with the actor and the argument, and explains the task map in plain terms as a description of AI R&D work with automation ratings. It preserves the stated purposes and caveats. The question follows from the stated next steps of adding tasks and re-rating as AI improves. A general reader can follow the central idea without the O*NET reference.","excerpt":"add tasks we missed, sharpen descriptions to be more precise and accurate, taxonomize to be more mutually exclusive and collectively exhaustive, and re-rate the list as AI improves"},{"dimension":"date","status":"pass","explanation":"evidence.originalPublished is 2026-06-17 at instant precision, and the reading states June 17, 2026, which matches the UTC day. The evidence kind is publisher-page, so no archive disclosure is required. The page mentions METR's most recent Frontier Risk Report, but the reading does not rely on it. There is no hindsight framing, and the authors' hopes are presented as hopes rather than outcomes.","excerpt":null},{"dimension":"support","status":"pass","explanation":"All claims are source-stated with exact excerpts. The statement that a task list alone cannot show AI performance per task is grounded in the authors' caveat. The stated uses (interpreting benchmarks, finding unmeasured tasks, studying uplift) match the text. No unsupported benefits or costs are implied.","excerpt":"Even if we’re armed with a perfectly comprehensive dataset of tasks, we’d still need to measure how AI performs on each task, for example."},{"dimension":"media","status":"pass","explanation":"No media was selected, and no verified media candidates were supplied, so a null media ID is appropriate.","excerpt":null}],"createdAt":"2026-10-30T06:24:06.972Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}