“Depressingly, despite all of the apparent progress in the last few years, much of what we raised in 2015 and re-emphasized in 2019 remains unsolved.”
Our reading, the sentence that was checked:
Much of the commonsense reasoning problem the authors raised in 2015 and 2019 remains unsolved in AI as of January 2025.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, AI still lacks “common” sense, 70 years later, paragraph 11.
- 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-b0c0ed-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)
Conditions
- despite apparent progress in the last few years (qualifier, stated by the speaker)
From the speaker
- much of what we raised in 2015
- re-emphasized in 2019
- remains unsolved
- despite all of the apparent progress
Our interpretation
- The ordinary reading is that substantial problems in reliable everyday reasoning persist, not that every example from the earlier publications still defeats AI.
- The earlier publications identify physical understanding, reasoning about people's knowledge and intentions, and applying knowledge reliably across situations.
- For the ongoing state described on January 5, I use 2024, the latest completed calendar year of publicly reportable research. This is an assessment window, not a claim that AI research follows an annual reporting cycle.
- Substantial failures across distinct relevant domains can support 'much'; I do not interpret it as a numerical majority of an unspecified problem list.
Counts as supported if: Independent evaluations in the target period establish substantial unresolved reliability problems across multiple commonsense domains identified in the earlier work.
Counts as contradicted if: Evidence establishes broadly reliable solutions across those domains, leaving only isolated or peripheral problems.
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
Independent 2024 evaluations document substantial failures in physical understanding and applied social reasoning. These match major concerns in the earlier work. Stronger performance on some mental-state tasks and improvements from guided prompting qualify the claim but do not establish that the broader reliability problems were solved. Support is for the ordinary qualitative reading of 'much,' not a numerical majority of historical examples.
Limitations
- The result depends on interpreting 'unsolved' as lacking broad reliability.
- Benchmark failures establish limits in tested systems, not impossibility for every AI approach.
- The complete 2019 book was not independently retested; its scope was checked through the supplied passage and the authors' contemporary interview.
Evidence (3 items)
Supports the claim: Zheng et al., ICML 2024; arXiv,
struggle with answering
Why it matters here: ContPhy finds persistent problems inferring physical properties and reasoning about dynamics and counterfactuals. These overlap the earlier work's physical-understanding concerns; they do not retest every historical example.
- Publisher
- Zheng et al., ICML 2024; arXiv
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Supports the claim: Gu et al.; arXiv,
judgment task.
Why it matters here: SimpleToM reports failures applying information about people's knowledge to behavior judgments, including a 59.5% judgment score for o1-preview. It also reports large improvements from specific prompting interventions, so the evidence concerns dependable application rather than complete absence of knowledge.
- Publisher
- Gu et al.; arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Street et al.; arXiv,
better-than-human performance
Why it matters here: This is material evidence of progress in reasoning about mental states. Success on this task does not establish reliable performance across physical and applied social reasoning.
- Publisher
- Street et al.; arXiv
- 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, AI still lacks “common” sense, 70 years later
Depressingly, despite all of the apparent progress in the last few years, much of what we raised in 2015 and re-emphasized in 2019 remains unsolved.
- Where
- paragraph 11
- Published
- 2025-01-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-ai-still-lacks-common-sense-70-years-b0c0ed-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.
About assessment marcus-ai-still-lacks-common-sense-70-years-b0c0ed-a1.