“this too seems like a rather large assumption that is far from proven, a possibility but not demonstrably a default,”
Our reading, the sentence that was checked:
Yudkowsky's claim that AGI will kill everyone by default is far from proven; it is possible but not shown to be the default.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, 𝐩(𝐝𝐨𝐨𝐦), footnote 2.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-d28-ea059a-a1
We recorded 24 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-29. The source may have been edited after 2023-08-27. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
From the speaker
- AGI will kill us all by default
- superintelligence will by default dominate us and harvest our atoms
- far from proven
- a possibility but not demonstrably a default
Our interpretation
- This concerns the public evidential status of the extinction argument when Marcus published, rather than whether extinction will eventually occur.
- The target period covers the public record through 2023-08-27. There is no natural reporting quarter for this observation about accumulated knowledge.
- The plain reading of 'default' is the expected outcome of developing superhuman AGI without a successful intervention to make it safe.
- 'Proven' means established by evidence or reasoning with substantiated premises. It does not require mathematical certainty or an actual extinction event.
- 'A possibility' concedes a plausible risk; it does not assign that risk a numerical probability.
Counts as supported if: The available arguments and research identify plausible routes to catastrophe but leave material assumptions unestablished between AI capabilities, power-seeking, and human extinction as the expected outcome.
Counts as contradicted if: Evidence available by publication establishes those material links sufficiently to demonstrate human extinction as the default outcome under the relevant conditions.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
On the stated plain reading, Marcus's distinction is supported. Yudkowsky presented a substantive argument for catastrophe, and formal research supplied reasons to expect power-seeking under specified conditions. But the examined sources did not establish the full chain from developing AGI to human extinction as the default. The power-seeking researchers expressly limited what their results demonstrated; their broader follow-up still described a possible safety risk. This is an assessment of the evidence available in August 2023, not a finding that extinction is unlikely or that Yudkowsky's forecast will fail.
Limitations
- The conclusion involves evaluating the sufficiency of arguments, not checking a numerical measurement.
- The absence of a mathematical proof alone would not support Marcus; the assessment also considers the unestablished links to real-world behavior and extinction.
- The source-type options lack a category for original argumentative essays; e1 and e2 are primary statements of Yudkowsky's reasoning.
Evidence (4 items)
Supports the claim: arXiv; Alexander Matt Turner and coauthors,
do not mathematically prove
Why it matters here: The authors explicitly limit the inference from their formal results to hypothetical superintelligent agents, noting differences between optimal and learned policies. This supports an evidential gap in a key step toward default extinction, although it does not by itself evaluate every possible extinction argument.
- Publisher
- arXiv; Alexander Matt Turner and coauthors
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: TIME; opinion essay by Eliezer Yudkowsky,
an opposed superhuman intelligence
Why it matters here: This primary opinion essay gives the argument Marcus addresses: inadequate alignment, hostile superhuman capabilities, and catastrophic defeat. It supplies reasons for concern, but the passage assumes the opposed superhuman system rather than establishing how often that situation arises.
- Publisher
- TIME; opinion essay by Eliezer Yudkowsky
- Kind of source
- News report
- Describes events of
- Public from
Context only: LessWrong; original argument by Eliezer Yudkowsky,
There are no pivotal weak acts.
Why it matters here: Yudkowsky argues that limited, relatively safe AI cannot prevent others from building dangerous systems. My assessment is that this remains a substantive premise of his extinction argument, rather than a demonstrated exclusion of all such interventions.
- Where
- Section A, point 7
- Publisher
- LessWrong; original argument by Eliezer Yudkowsky
- Kind of source
- Secondary summary
- Describes events of
- Public from
Context only: arXiv; Alexander Matt Turner and Prasad Tadepalli,
These results suggest a safety risk:
Why it matters here: This follow-up extends the theory beyond optimal policies, so the earlier limitations must not be mistaken for an absence of supporting research. Its stated implication is a possible real-world power-seeking risk, not a demonstrated default of human extinction.
- Publisher
- arXiv; Alexander Matt Turner and Prasad Tadepalli
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, 𝐩(𝐝𝐨𝐨𝐦)
this too seems like a rather large assumption that is far from proven, a possibility but not demonstrably a default,
- Where
- footnote 2
- Published
- 2023-08-27
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-d28-ea059a-a1: Current. Outcome: Supported. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.