Ajeya Cotra · Mar 26, 2023
Ajeya Cotra argues alignment researchers need clearer shared assumptions
Markdown · JSON · Original source
Ajeya Cotra’s article was published on March 26, 2023. Cotra describes AI alignment research as technical work on building powerful AI systems that try to do what their designers intend, rather than deliberately circumventing or disregarding that intent. She says the field lacks a shared foundation, in part because researchers disagree about how to define alignment, how likely misalignment is, what its consequences might be, and which research matters most. Cotra also emphasizes that researchers may hold very different assumptions about the risks of advanced AI and the value of different approaches. She describes her own view: that advanced AI could pose severe risks, and that alignment research can make progress within the deep learning paradigm. She notes that many, but not all, researchers share aspects of that perspective. Cotra argues that researchers could work more effectively if they developed common assumptions, or at least made their existing assumptions clearer. She is cautiously optimistic about that possibility, while describing it as a long and difficult process. For people new to the field, her central point is that research from one camp may be unintelligible to another because they start from different assumptions. What shared assumptions, if any, could alignment researchers agree on without obscuring their substantive disagreements?
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
Ajeya Cotra · Article
Source details
Original source date: Mar 26, 2023 (2023-03-26T18:08:18+00:00).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-10-29T15:59:57.726Z. First observed: 2026-10-28T20:54:01.566Z. Retrieved: 2026-10-28T20:55:34.470Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on October 28, 2026; originally published March 26, 2023. Cotra presents her perspective and acknowledges that researchers disagree; the article does not establish how representative any camp or position is. The article describes possible risks and research priorities, not demonstrated outcomes of future AI systems.
Open question
What shared assumptions, if any, could alignment researchers agree on without obscuring their substantive disagreements?
Source-stated
Cotra defines AI alignment research as technical work intended to make powerful AI systems act in accordance with their designers’ intentions.
technical research that’s trying to develop ways of building powerful AI systems so that they “try their best” to do what their designers want them to do
Source-stated
Cotra says researchers disagree about the definition, likelihood, consequences, and priorities of AI alignment research.
vastly different ideas about how to define the alignment problem, how likely misalignment is in the first place, what the consequences of powerful misaligned AI are likely to be
Source-stated
Cotra argues that a shared set of assumptions, or clearer articulation of existing assumptions, could help the field work more effectively and make disagreements easier to understand.
the field as a whole could be much more effective if different camps could arrive at a workable set of common assumptions
Source-stated
Cotra says her own concern about alignment includes the possibility that advanced AI systems could take over the world, while noting that not all relevant researchers share her view.
I think that there’s a good chance that advanced AI systems will take over the world in a possibly-violent uprising or coup unless we make more research progress; many but not all people working on the relevant technical research problems share this picture.
Source and contributor context
Ajeya Cotra · 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.
AI alignment
Research and practice aimed at making AI systems pursue the goals their builders and users intend.
Deep learning
Machine learning that uses many layers of simple calculations; the basis of most current AI.
Editorial provenance
Version 1. Publication ID: d49ff4e8-fc81-7b31-b101-ff15397f9de4.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T15:58:33.696Z. 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: 41a391e5f469019f11fe423ff7576318e1c86559aa597ec426fb4716f5f40a79. Source text hash: 95520bd3d677687829f76342c3f9c0f666c91bcdc64d43d9719d24a4888aab24.
Revision and source-check history
Version 1 · 2026-10-29T15:59:57.726Z
Ajeya Cotra argues alignment researchers need clearer shared assumptions
Ajeya Cotra’s article was published on March 26, 2023. Cotra describes AI alignment research as technical work on building powerful AI systems that try to do what their designers intend, rather than deliberately circumventing or disregarding that intent. She says the field lacks a shared foundation, in part because researchers disagree about how to define alignment, how likely misalignment is, what its consequences might be, and which research matters most. Cotra also emphasizes that researchers may hold very different assumptions about the risks of advanced AI and the value of different approaches. She describes her own view: that advanced AI could pose severe risks, and that alignment research can make progress within the deep learning paradigm. She notes that many, but not all, researchers share aspects of that perspective. Cotra argues that researchers could work more effectively if they developed common assumptions, or at least made their existing assumptions clearer. She is cautiously optimistic about that possibility, while describing it as a long and difficult process. For people new to the field, her central point is that research from one camp may be unintelligible to another because they start from different assumptions. What shared assumptions, if any, could alignment researchers agree on without obscuring their substantive disagreements?
Original source
Ajeya Cotra · Article
Source details
Original source date: Mar 26, 2023 (2023-03-26T18:08:18+00:00).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-10-29T15:59:57.726Z. First observed: 2026-10-28T20:54:01.566Z. Retrieved: 2026-10-28T20:55:34.470Z.
Limits
Text as retrieved on October 28, 2026; originally published March 26, 2023. Cotra presents her perspective and acknowledges that researchers disagree; the article does not establish how representative any camp or position is. The article describes possible risks and research priorities, not demonstrated outcomes of future AI systems.
Open question
What shared assumptions, if any, could alignment researchers agree on without obscuring their substantive disagreements?
Source-stated
Cotra defines AI alignment research as technical work intended to make powerful AI systems act in accordance with their designers’ intentions.
technical research that’s trying to develop ways of building powerful AI systems so that they “try their best” to do what their designers want them to do
Source-stated
Cotra says researchers disagree about the definition, likelihood, consequences, and priorities of AI alignment research.
vastly different ideas about how to define the alignment problem, how likely misalignment is in the first place, what the consequences of powerful misaligned AI are likely to be
Source-stated
Cotra argues that a shared set of assumptions, or clearer articulation of existing assumptions, could help the field work more effectively and make disagreements easier to understand.
the field as a whole could be much more effective if different camps could arrive at a workable set of common assumptions
Source-stated
Cotra says her own concern about alignment includes the possibility that advanced AI systems could take over the world, while noting that not all relevant researchers share her view.
I think that there’s a good chance that advanced AI systems will take over the world in a possibly-violent uprising or coup unless we make more research progress; many but not all people working on the relevant technical research problems share this picture.
Source and contributor context
Ajeya Cotra · 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.
AI alignment
Research and practice aimed at making AI systems pursue the goals their builders and users intend.
Deep learning
Machine learning that uses many layers of simple calculations; the basis of most current AI.
Editorial provenance
Version 1. Publication ID: d49ff4e8-fc81-7b31-b101-ff15397f9de4.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T15:58:33.696Z. 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: 41a391e5f469019f11fe423ff7576318e1c86559aa597ec426fb4716f5f40a79. Source text hash: 95520bd3d677687829f76342c3f9c0f666c91bcdc64d43d9719d24a4888aab24.
passed · 2026-10-29T15:58:33.696Z · anthropic/claude-opus-5.5
Publication: d49ff4e8-fc81-7b31-b101-ff15397f9de4
attribution — pass: The reading correctly attributes the article and its views to Ajeya Cotra, the author shown on the publisher page. Cotra's personal views are framed as hers, and the reading notes that other researchers disagree.
Like anyone else who thinks about this area, I’m coming from my own particular perspective
summary — pass: The summary accurately conveys the definition of alignment research, the field's lack of shared foundations, the dimensions of disagreement, Cotra's own view, and her call for common or clearer assumptions. Alignment is explained in plain language for a general reader. Describing her takeover concern as 'severe risks' is softer than the source but not misleading, and claim 4 states it directly. The question follows from the source without leading.
It’s a pre-paradigmatic field — there’s very little shared foundation that researchers can lay bricks on top of right now.
date — pass: The stated date, March 26, 2023, matches evidence.originalPublished (2023-03-26T18:08 UTC) and the dateline. modifiedAt is the same instant, so there is no sign of a later edit. The evidence kind is publisher-page, so no archive disclosure is needed. The reading contains no hindsight framing and presents nothing from after that date as known.
Ajeya Cotra Mar 26, 2023
support — pass: All four claims are source-stated, and each excerpt matches the text verbatim. The limitations correctly frame the article's risks as possibilities and her position as one perspective among several. The reading draws no unsupported inferences about benefits or costs.
I think that the field as a whole could be much more effective if different camps could arrive at a workable set of common assumptions that could be built upon.
media — pass: No media was selected and no verified media candidates were supplied, so selecting none is appropriate.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.