“In the real world, only one theorem of middling importance (the Robbins conjecture) has ever been proved autonomously by an AI system, and that was back in 1996.”
Our reading, the sentence that was checked:
Only one theorem of middling importance (the Robbins conjecture, 1996) has ever been proved autonomously by an AI system.
Assessment: Contradicted Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, DeepMind and OpenAI achieve IMO Gold. What does it all mean?, paragraph 22.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-deepmind-and-openai-achieve-imo-gold-82c1f1-a1
We recorded 15 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2025-07-22. 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
- only one theorem of middling importance
- the Robbins conjecture
- has ever been proved autonomously by an AI system
- back in 1996
Our interpretation
- The historical cutoff is July 22, 2025; 'ever' leaves the beginning unbounded.
- In context, the claim excludes other autonomous proofs of original research results of at least meaningful mathematical importance, not merely exercises or reproving known theorems.
- Because Marcus counts EQP, 'AI system' includes symbolic automated reasoning systems and is not restricted to neural networks.
- Autonomous proof search permits humans to state and encode the problem, configure the search, and check the result. It excludes humans supplying the substantive proof for a computer merely to verify.
- Solutions of longstanding recognized research problems are treated as meeting the ordinary qualitative importance threshold. This does not assert an exact ranking against Robbins.
Counts as supported if: A sufficiently comprehensive historical record establishes Robbins as the only qualifying original research theorem proved by automated reasoning through the cutoff.
Counts as contradicted if: At least one additional nontrivial, recognized open research problem was settled by autonomous proof search before July 22, 2025, with a validated proof.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Contradicted
The dated evidence goes against 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
Robbins was proved automatically in 1996, but it was not the only meaningful research result obtained by automated proof search before July 2025. Boolean Pythagorean triples and Schur number five provide documented counterexamples with validated proofs. This judgment counts human problem encoding and search configuration as compatible with autonomy, consistent with the treatment of EQP.
Limitations
- Mathematical importance is qualitative; the sources do not supply an exact comparison with Robbins.
- These were symbolic reasoning systems with substantial human setup.
- A narrower definition excluding SAT solvers or requiring no human setup would change the assessment and would need explicit justification.
Evidence (3 items)
Goes against the claim: EasyChair, GCAI 2017; Marijn Heule,
completely automatically
Why it matters here: The paper identifies the Boolean Pythagorean triples problem as its example and documents a generated proof validated using formally verified checkers, including independent efforts. Its 2016 solution supplies another longstanding research problem settled through automated proof search.
- Publisher
- EasyChair, GCAI 2017; Marijn Heule
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Goes against the claim: arXiv; Marijn J. H. Heule,
validated a proof
Why it matters here: The paper reports solving the century-old Schur number five problem by encoding it into propositional logic and applying parallel SAT solving. It gives the answer as 160 and reports certification with a formally verified proof checker.
- Where
- Abstract
- Publisher
- arXiv; Marijn J. H. Heule
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Argonne National Laboratory; William McCune,
found automatically by EQP
Why it matters here: This confirms the substance of the Robbins example. The paper also describes repeated human selection of search parameters, showing that autonomy here cannot mean absence of human setup.
- Publisher
- Argonne National Laboratory; William McCune
- 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, DeepMind and OpenAI achieve IMO Gold. What does it all mean?
In the real world, only one theorem of middling importance (the Robbins conjecture) has ever been proved autonomously by an AI system, and that was back in 1996.
- Where
- paragraph 22
- Published
- 2025-07-22
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-deepmind-and-openai-achieve-imo-gold-82c1f1-a1: Current. Outcome: Contradicted. 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.
About assessment marcus-deepmind-and-openai-achieve-imo-gold-82c1f1-a1.