# Gabriel Unger and Arvind Narayanan argue in Normal Tech for AI beyond automation [Canonical reading](https://escapemeridian.com/readings/gabriel-unger-and-arvind-narayanan-argue-in-normal-tech-for-ai-5a4e1d4b) Originally published October 14, 2026, Gabriel Unger and Arvind Narayanan argue that AI research is too heavily oriented toward automating human labor, and should also pursue systems that help people work with AI. They do not argue that automation is always bad: they say some work is well suited to it, while other work could benefit more from collaboration. Their proposed research directions include systems that show people their intermediate steps, respond quickly enough to support interaction, let people and AI work on shared tasks continuously, and assess how well they work together. The authors connect these ideas to human stakes: they argue that work matters to people beyond receiving income, and that AI could help people do new things rather than only automate existing tasks. They also say the automation-focused approach may struggle economically, while acknowledging that the alternative remains a minority position and needs a systematic research program. Their proposed route is bottom-up: technical experimentation, changes in computer science, targeted policy, and firms discovering reasons to pursue collaboration. The research directions are explicitly tentative and experimental; the authors’ hope is that collective effort can steer AI without top-down commands. Whether this alternative can produce useful systems and wider support remains open. 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. ## Questions using this source These authored EM reads cite this source. Follow their assumptions and limitations; inclusion does not imply agreement. - [Who gets an entry point and shares the gains?](/questions/ai-entry-and-worker-gains) ## Original source Arvind Narayanan and Sayash Kapoor · Article Original source date: Oct 14, 2026 (2026-10-14T19:56:44.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-11-04T22:33:24.195Z. First observed: 2026-10-23T12:55:24.923Z. Retrieved: 2026-10-23T13:38:51.395Z. - [Steering AI Beyond Automation](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## Limits Text as retrieved on October 23, 2026; originally published October 14, 2026. The authors describe the research directions as tentative and experimental; the source does not establish that they have produced working collaboration agents or demonstrated benefits for workers. ## Open question What evidence would show that human-AI collaboration produces useful results for workers and firms, rather than merely shifting how tasks are performed? ## Source-stated Gabriel Unger and Arvind Narayanan argue that AI development is too focused on automation and should also pursue human-AI collaboration. > right now too much of AI research is devoted to the latter at the expense of the former. - [Supporting source](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## Source-stated The authors propose research into agents that reveal intermediate processes, interact with low latency, support continuous collaboration, and are evaluated with humans. > We present four such illustrative directions. - [Supporting source](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## Source-stated The authors argue that automation is not universally bad, but that some work is better served by augmentation and collaboration. > Automation is not a universally bad thing. Many jobs are extremely routine and naturally amenable to automation. - [Supporting source](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## Source-stated The authors propose a bottom-up approach and hope collective effort can steer AI development without top-down commands. > Our hope is that through greater collective awareness and intention, and better pooling and coordination of the many efforts that already exist, a bottom-up, decentralized approach can be successful at steering the direction of AI development and deployment, without requiring top-down commands or - [Supporting source](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## Source-stated The authors describe their proposed research directions as tentative and experimental. > The specific directions we sketch are intended to be tentative and experimental. - [Supporting source](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## 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. ## Editorial provenance Version 1. Publication ID: dc92bb16-2dc0-5c63-2ec8-627fc8e54c5a. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-04T22:28:42.817Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_screen. Observation: 77b0f8b00b97c585d583b987be6c122909aed094de50fc550c5297b64af45a5b. Source text hash: f2b6cc0e1c8b6bbfe0e9efb6e94a5614b87630de3934d9902943551c31faf389. ## Revision and source-check history ### Version 1 — 2026-11-04T22:33:24.195Z Originally published October 14, 2026, Gabriel Unger and Arvind Narayanan argue that AI research is too heavily oriented toward automating human labor, and should also pursue systems that help people work with AI. They do not argue that automation is always bad: they say some work is well suited to it, while other work could benefit more from collaboration. Their proposed research directions include systems that show people their intermediate steps, respond quickly enough to support interaction, let people and AI work on shared tasks continuously, and assess how well they work together. The authors connect these ideas to human stakes: they argue that work matters to people beyond receiving income, and that AI could help people do new things rather than only automate existing tasks. They also say the automation-focused approach may struggle economically, while acknowledging that the alternative remains a minority position and needs a systematic research program. Their proposed route is bottom-up: technical experimentation, changes in computer science, targeted policy, and firms discovering reasons to pursue collaboration. The research directions are explicitly tentative and experimental; the authors’ hope is that collective effort can steer AI without top-down commands. Whether this alternative can produce useful systems and wider support remains open. ## Questions using this source These authored EM reads cite this source. Follow their assumptions and limitations; inclusion does not imply agreement. - [Who gets an entry point and shares the gains?](/questions/ai-entry-and-worker-gains) ## Original source Arvind Narayanan and Sayash Kapoor · Article Original source date: Oct 14, 2026 (2026-10-14T19:56:44.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-11-04T22:33:24.195Z. First observed: 2026-10-23T12:55:24.923Z. Retrieved: 2026-10-23T13:38:51.395Z. - [Steering AI Beyond Automation](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## Limits Text as retrieved on October 23, 2026; originally published October 14, 2026. The authors describe the research directions as tentative and experimental; the source does not establish that they have produced working collaboration agents or demonstrated benefits for workers. ## Open question What evidence would show that human-AI collaboration produces useful results for workers and firms, rather than merely shifting how tasks are performed? ## Source-stated Gabriel Unger and Arvind Narayanan argue that AI development is too focused on automation and should also pursue human-AI collaboration. > right now too much of AI research is devoted to the latter at the expense of the former. - [Supporting source](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## Source-stated The authors propose research into agents that reveal intermediate processes, interact with low latency, support continuous collaboration, and are evaluated with humans. > We present four such illustrative directions. - [Supporting source](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## Source-stated The authors argue that automation is not universally bad, but that some work is better served by augmentation and collaboration. > Automation is not a universally bad thing. Many jobs are extremely routine and naturally amenable to automation. - [Supporting source](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## Source-stated The authors propose a bottom-up approach and hope collective effort can steer AI development without top-down commands. > Our hope is that through greater collective awareness and intention, and better pooling and coordination of the many efforts that already exist, a bottom-up, decentralized approach can be successful at steering the direction of AI development and deployment, without requiring top-down commands or - [Supporting source](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## Source-stated The authors describe their proposed research directions as tentative and experimental. > The specific directions we sketch are intended to be tentative and experimental. - [Supporting source](https://www.normaltech.ai/p/steering-ai-beyond-automation) ## 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. ## Editorial provenance Version 1. Publication ID: dc92bb16-2dc0-5c63-2ec8-627fc8e54c5a. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-04T22:28:42.817Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_screen. Observation: 77b0f8b00b97c585d583b987be6c122909aed094de50fc550c5297b64af45a5b. Source text hash: f2b6cc0e1c8b6bbfe0e9efb6e94a5614b87630de3934d9902943551c31faf389. ### passed — 2026-11-04T22:28:42.817Z — anthropic/claude-opus-5.5 Publication: dc92bb16-2dc0-5c63-2ec8-627fc8e54c5a attribution — pass: The byline names Gabriel Unger and Arvind Narayanan as the authors. The reading attributes the arguments to them and keeps them as the authors' views. The source is a single essay with its own argument, not a listing. > By Gabriel Unger and Arvind Narayanan. Unger is an economist at Stanford’s Digital Economy Lab. summary — pass: The reading captures the core thesis, the authors' caveat that automation is sometimes good, the four research directions (described in plain language), the economic and human stakes, and the bottom-up route. Central ideas are understandable without jargon. It keeps the authors' conviction while noting the alternative is a minority position. The question follows from the evidence and is not leading. > This view — our view — is not that automation is never good. date — pass: The instant 2026-10-14T19:56Z falls on the UTC day of October 14, 2026, which matches the source dateline. The reading uses no later events and no hindsight framing. It presents the authors' hopes and predictions as their own. The page is a publisher page, so no archive disclosure is needed. > Arvind Narayanan Oct 14, 2026 support — pass: Each claim is backed by a source excerpt. The point that work matters beyond income rests on the authors' statement about real work. The struggle-economically point rests on the trilemma and ROI discussion. The limitations correctly note that the directions are tentative and that no benefits have been demonstrated. No stakeholder impacts are invented. > Most Americans want real lives with real work, not welfare checks. media — pass: No media was selected, and no verified media candidates were supplied, so no media identity or relevance issue arises. ## 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/5a4e1d4b2fb65d2a40b74bca91665b9dc8321e961ea8e51dea6a078d98139bff/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/gabriel-unger-and-arvind-narayanan-argue-in-normal-tech-for-ai-5a4e1d4b.json)