· Oct 6, 2026
Scott Alexander argues in Astral Codex Ten that Pinker misrepresents AI risk
Markdown · JSON · Original source
’s letter was published on October 6, 2026. He answers Steven Pinker’s arguments against and accuses Pinker of portraying AI risk advocates as fatalistic, imagining omniscient AI, and dismissing present harms. Alexander argues that AI could develop self-preservation or power-seeking through convergent instrumental goals, misgeneralization, or , without natural selection. He also argues that systems with multiple goals could still harm people and other beings whose preferences differ from theirs. Alexander’s central grievance is that Pinker’s public framing, in his view, can lead readers to dismiss AI risk advocates before considering their arguments. He points to cooperation between long-term-risk and near-term-harm groups as a response to the claim that existential-risk concerns distract from present harms. He concludes by urging Pinker to confront risks through research and preparation, rather than denial or blind panic. Alexander predicts that public recognition of AI’s danger may force a choice between investing in and banning smarter-than-human AI. He worries that Pinker may discredit techno-optimism before it is needed.
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
· Article
Source details
Original source date: Oct 6, 2026 (2026-10-06T21:23:06.000Z).
Date basis: Date supplied by the publisher’s feed.
Added to Escape Meridian: 2026-11-07T04:33:26.237Z. First observed: 2026-10-23T23:10:02.541Z. Retrieved: 2026-10-23T23:11:37.591Z.
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 23, 2026; originally published October 6, 2026. The letter presents Alexander’s account of Pinker’s arguments and of the AI risk community; it does not include Pinker’s full response or independent verification of the disputed characterizations.
Open question
What evidence would show whether Alexander’s account of Pinker’s claims and of cooperation between AI-risk groups is representative?
Source-stated
Alexander argues that AI could develop self-preservation or power-seeking through mechanisms that do not depend on natural selection.
There are approximately three reasons why we think AI might have self-preservation and dominance drives, and none of them involve natural selection. First is convergent instrumental goals, second is misgeneralization, and third is reward-hacking.
Source-stated
Alexander argues that systems with multiple goals could still harm people or beings whose preferences differ from theirs.
Everyone else with different preferences for flourishing - like Manhattan Island wildlife who prefer quiet forests to bustling cities - is equally screwed either way.
Source-stated
Alexander says Pinker’s public framing can lead readers to dismiss AI risk advocates before hearing their arguments.
These people will listen to you and conclude that “doomers” are hopeless pessimists who are “certain of” a “100% chance” of doom, “fatalistic” about it, and think we should lie down and wait to die.
Source-stated
Alexander presents cooperation between long-term-risk and near-term-harm groups as a response to the claim that existential-risk concerns distract from present harms.
In the end, the Senate went from broadly in favor to rejecting it 99-1.
Source-stated
Alexander argues for addressing AI risks through research and preparation, rather than denial or blind panic.
We should face the danger with clear eyes, trusting the human faculties of reason, optimism, and compassion to bring us through.
Source-stated
Alexander predicts that public recognition of AI’s danger may force a choice between investing in alignment research and banning smarter-than-human AI.
Once it becomes obvious to the average person that AI really is a big deal and really is dangerous - and I am no longer sure this moment is in the future - we will have to choose between two visions.
Source and contributor context
Scott Alexander · 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.
Superintelligence
A hypothetical AI far more capable than people across most intellectual tasks.
Reward hacking
When an AI finds a shortcut that scores well on its training objective without doing what its designers intended.
AI alignment
Research and practice aimed at making AI systems pursue the goals their builders and users intend.
Scott Alexander
Scott Alexander authored essays published in the Slate Star Codex archive.
Editorial provenance
Version 1. Publication ID: cfa17765-cf23-70a2-435f-9b1d63416e95.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-07T04:29:03.163Z. 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: c8028d53af6dd8ef03e9044882178ea07aabbbb0af3f9e837a20e9b3130441ba. Source text hash: c06a6d47668e7e58a9caeeb3c57ecbc9a2422f4780607943159cae81c1f30bc1.
Revision and source-check history
Version 1 · 2026-11-07T04:33:26.237Z
Scott Alexander argues in Astral Codex Ten that Pinker misrepresents AI risk
’s letter was published on October 6, 2026. He answers Steven Pinker’s arguments against and accuses Pinker of portraying AI risk advocates as fatalistic, imagining omniscient AI, and dismissing present harms. Alexander argues that AI could develop self-preservation or power-seeking through convergent instrumental goals, misgeneralization, or , without natural selection. He also argues that systems with multiple goals could still harm people and other beings whose preferences differ from theirs. Alexander’s central grievance is that Pinker’s public framing, in his view, can lead readers to dismiss AI risk advocates before considering their arguments. He points to cooperation between long-term-risk and near-term-harm groups as a response to the claim that existential-risk concerns distract from present harms. He concludes by urging Pinker to confront risks through research and preparation, rather than denial or blind panic. Alexander predicts that public recognition of AI’s danger may force a choice between investing in and banning smarter-than-human AI. He worries that Pinker may discredit techno-optimism before it is needed.
Original source
· Article
Source details
Original source date: Oct 6, 2026 (2026-10-06T21:23:06.000Z).
Date basis: Date supplied by the publisher’s feed.
Added to Escape Meridian: 2026-11-07T04:33:26.237Z. First observed: 2026-10-23T23:10:02.541Z. Retrieved: 2026-10-23T23:11:37.591Z.
Limits
Text as retrieved on October 23, 2026; originally published October 6, 2026. The letter presents Alexander’s account of Pinker’s arguments and of the AI risk community; it does not include Pinker’s full response or independent verification of the disputed characterizations.
Open question
What evidence would show whether Alexander’s account of Pinker’s claims and of cooperation between AI-risk groups is representative?
Source-stated
Alexander argues that AI could develop self-preservation or power-seeking through mechanisms that do not depend on natural selection.
There are approximately three reasons why we think AI might have self-preservation and dominance drives, and none of them involve natural selection. First is convergent instrumental goals, second is misgeneralization, and third is reward-hacking.
Source-stated
Alexander argues that systems with multiple goals could still harm people or beings whose preferences differ from theirs.
Everyone else with different preferences for flourishing - like Manhattan Island wildlife who prefer quiet forests to bustling cities - is equally screwed either way.
Source-stated
Alexander says Pinker’s public framing can lead readers to dismiss AI risk advocates before hearing their arguments.
These people will listen to you and conclude that “doomers” are hopeless pessimists who are “certain of” a “100% chance” of doom, “fatalistic” about it, and think we should lie down and wait to die.
Source-stated
Alexander presents cooperation between long-term-risk and near-term-harm groups as a response to the claim that existential-risk concerns distract from present harms.
In the end, the Senate went from broadly in favor to rejecting it 99-1.
Source-stated
Alexander argues for addressing AI risks through research and preparation, rather than denial or blind panic.
We should face the danger with clear eyes, trusting the human faculties of reason, optimism, and compassion to bring us through.
Source-stated
Alexander predicts that public recognition of AI’s danger may force a choice between investing in alignment research and banning smarter-than-human AI.
Once it becomes obvious to the average person that AI really is a big deal and really is dangerous - and I am no longer sure this moment is in the future - we will have to choose between two visions.
Source and contributor context
Scott Alexander · 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.
Superintelligence
A hypothetical AI far more capable than people across most intellectual tasks.
Reward hacking
When an AI finds a shortcut that scores well on its training objective without doing what its designers intended.
AI alignment
Research and practice aimed at making AI systems pursue the goals their builders and users intend.
Scott Alexander
Scott Alexander authored essays published in the Slate Star Codex archive.
Editorial provenance
Version 1. Publication ID: cfa17765-cf23-70a2-435f-9b1d63416e95.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-07T04:29:03.163Z. 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: c8028d53af6dd8ef03e9044882178ea07aabbbb0af3f9e837a20e9b3130441ba. Source text hash: c06a6d47668e7e58a9caeeb3c57ecbc9a2422f4780607943159cae81c1f30bc1.
uncertain · 2026-11-07T04:29:03.163Z · anthropic/claude-opus-5.5
Publication: cfa17765-cf23-70a2-435f-9b1d63416e95
attribution — pass: The reading attributes the letter and its arguments to Scott Alexander, who publishes Astral Codex Ten, at the matching URL. It is one argumentative piece, not a listing, and Pinker's positions are presented as Alexander's characterizations.
I have the honor to be your obedient servant, Scott Alexander
summary — uncertain: The summary is broadly faithful, but it names 'convergent instrumental goals' and 'misgeneralization' as central mechanisms without explaining them. Neither appears in the definitions, which cover only superintelligence, reward hacking and alignment. A general reader cannot see the source's mechanism. Smallest repair: briefly gloss both terms, for example goals that help almost any task, such as staying alive. The reading also never identifies Steven Pinker and flattens the letter's combative force: the duel joke and the charge of dishonorable writing.
First is convergent instrumental goals, second is misgeneralization, and third is reward-hacking.
date — pass: evidence.originalPublished is 2026-10-06T21:23Z. The reading's October 6, 2026 matches both that UTC day and the source dateline. The retrieval date is stated separately. Alexander's prediction is framed as a prediction, with no hindsight framing. modifiedAt equals the publication time, and the source is a publisher page, so no archive disclosure is needed.
Scott Alexander Oct 06, 2026
support — pass: Each claim is source-stated and supported by its quoted text. These include the three non-evolutionary mechanisms, harm from multi-goal systems, well-poisoning, the coalition against preemption, facing danger with resolution, and the two-visions choice. The worry about discrediting techno-optimism is also stated in the source. The question is genuinely open and not leading.
But I worry that you will have discredited yourself - and, by extension, poisoned the brand of techno-optimism you promote - right before the point where it becomes most necessary.
media — pass: No media was selected, and no verified media candidates were supplied, so there is no identity or relevance issue.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
Scott Alexander authored essays published in the Slate Star Codex archive.
Explore Scott AlexanderScott Alexander authored essays published in the Slate Star Codex archive.
Explore Scott AlexanderA hypothetical AI far more capable than people across most intellectual tasks.
Definition referenceWhen an AI finds a shortcut that scores well on its training objective without doing what its designers intended.
Definition referenceResearch and practice aimed at making AI systems pursue the goals their builders and users intend.
Definition referenceScott Alexander authored essays published in the Slate Star Codex archive.
Explore Scott AlexanderScott Alexander authored essays published in the Slate Star Codex archive.
Explore Scott AlexanderA hypothetical AI far more capable than people across most intellectual tasks.
Definition referenceWhen an AI finds a shortcut that scores well on its training objective without doing what its designers intended.
Definition referenceResearch and practice aimed at making AI systems pursue the goals their builders and users intend.
Definition referenceScott Alexander authored essays published in the Slate Star Codex archive.
Explore Scott Alexander