“Three approaches to commonsense reasoning have become popular in the last fifteen years; none has proved fully satisfactory.”
Our reading, the sentence that was checked:
None of the three popular approaches to commonsense reasoning (simulation, LLM emergence, video generation) has proved fully satisfactory.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, AI still lacks “common” sense, 70 years later, paragraph 13.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-ai-still-lacks-common-sense-70-years-4f1e61-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 2025-01-05. 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
- Three approaches to commonsense reasoning
- in the last fifteen years
- none has proved fully satisfactory
Our interpretation
- The three approaches are physical simulation, commonsense emerging in foundation models, and video generation.
- 'Fully satisfactory' means demonstrated, broadly reliable commonsense reasoning across ordinary situations. It does not require error-free performance.
- The stated fifteen-year interval ends on the post date.
- Success on a restricted task is relevant progress but does not by itself settle whether an entire approach has supplied a broad solution.
Counts as supported if: Evidence establishes material unresolved scope or reliability limits for all three approaches as of the post.
Counts as contradicted if: At least one approach has demonstrated broadly reliable commonsense reasoning by the post date.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
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
The evidence supports significant limitations in tested language models and video generators. Simulation research also shows meaningful successes. The retrieved studies do not establish the full three-approach claim as of January 2025, particularly its simulation component. Nor do the successful specialized systems contradict the ordinary reading that no broad solution had been demonstrated.
Limitations
- Partial support for two approaches is insufficient to score the entire conjunction supported.
- An old benchmark's failures cannot establish the contemporary limits of a whole approach.
Evidence (4 items)
Supports the claim: Association for Computational Linguistics,
consistently low
Why it matters here: The tested language models have substantial spatial-reasoning weaknesses, supporting the foundation-model part. The paper also reports improvements with scale and fine-tuning; available_at conservatively uses the end of the stated publication month.
- Where
- Abstract
- Publisher
- Association for Computational Linguistics
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Supports the claim: Meng et al.; arXiv,
physical commonsense.
Why it matters here: PhyGenBench finds physical-consistency failures across its tested video generators. This supports the video-generation part, within the models and physical laws tested.
- Where
- Abstract
- Publisher
- Meng et al.; arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Bakhtin et al.; arXiv,
fall short
Why it matters here: This establishes historical difficulty in a simulated physical-reasoning environment. It does not establish that simulation-based reasoning remained broadly inadequate five years later, or that the simulator itself was the failing component.
- Where
- Abstract
- Publisher
- Bakhtin et al.; arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Zhu et al., ICLR 2024; arXiv,
strong performance
Why it matters here: The model infers hidden fluid properties and transfers them to new simulated scenes. This is meaningful progress under incomplete physical information, but its restricted task scope does not establish a general commonsense solution.
- Publisher
- Zhu et al., ICLR 2024; arXiv
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
What would settle it
A contemporaneous synthesis or reproducible evaluations covering leading systems in all three approaches, with explicit evidence about broad reliability and generalization.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, AI still lacks “common” sense, 70 years later
Three approaches to commonsense reasoning have become popular in the last fifteen years; none has proved fully satisfactory.
- Where
- paragraph 13
- Published
- 2025-01-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-ai-still-lacks-common-sense-70-years-4f1e61-a1: Current. Outcome: Unresolved. 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-ai-still-lacks-common-sense-70-years-4f1e61-a1.