“(Spoiler alert: they’re not, at least not yet.)”
Our reading, the sentence that was checked:
Current approaches to getting machines to reason about causality are not adequate, at least not yet.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, An epic AI Debate—and why everyone should be at least a little bit worried about AI going into 2023, paragraph 6.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-an-epic-ai-debateand-why-everyone-90c73e-a1
We recorded 7 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 2022-12-31. 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
- at least not yet (qualifier, stated by the speaker)
From the speaker
- any of our current approaches
- getting machines to reason about causality
- adequate
- they’re not, at least not yet
Our interpretation
- The ordinary reading is that existing approaches do not yet provide dependable, broadly useful causal reasoning. This does not mean they cannot solve any causal problem.
- I assess reasoning about causal relationships, interventions, and counterfactuals, including specialized causal methods as well as language models.
- The ongoing state described is December 2022, the month of the debate and post. Earlier experiments provide context unless their findings establish performance during that month.
- Marcus supplies neither an adequacy threshold nor a specified application. Broad reliability is my interpretation, not an explicit benchmark requirement.
Counts as supported if: Evaluations applicable to December 2022 establish substantial reliability or generalization failures across the relevant approaches, sufficient to support the broad claim of inadequacy.
Counts as contradicted if: An approach available in December 2022 demonstrably provides dependable causal reasoning across varied relevant problems, including interventions and counterfactuals, under clearly stated assumptions.
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 research supports concern about unreliable causal answers in language models. It also documents successful specialist causal methods. These findings do not settle whether existing approaches collectively were inadequate in December 2022. Calling the claim supported would require generalizing from limited evaluations; calling it contradicted would require treating specialist success as sufficient for an unstated standard of adequacy.
Limitations
- The cited experiments and theoretical results are historical context, not comprehensive measurements within the target month.
- A narrower claim about language-model reliability would be easier to assess than this claim about current approaches generally.
Evidence (2 items)
Context only: arXiv; Moritz Willig and colleagues,
changing the query wording.
Why it matters here: The study reports unstable causal-graph answers, but also GPT-3 successes on 11 of 15 abstract questions and 21 of 36 intuitive-physics questions. Its limited model coverage and unspecified experiment dates prevent treating it as a comprehensive December 2022 outcome test.
- Publisher
- arXiv; Moritz Willig and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Kevin Xia and colleagues,
Simulations corroborate the proposed approach.
Why it matters here: The paper develops an algorithm for determining whether causal effects are identifiable and estimating them when they are, given structural constraints. This challenges a reading that existing approaches were uniformly ineffective, but does not demonstrate broad autonomous causal reasoning in December 2022.
- Publisher
- arXiv; Kevin Xia and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A representative historical evaluation of December 2022 approaches across causal relationships, interventions, and counterfactuals, with an adequacy standard justified independently of the results.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, An epic AI Debate—and why everyone should be at least a little bit worried about AI going into 2023
(Spoiler alert: they’re not, at least not yet.)
- Where
- paragraph 6
- Published
- 2022-12-31
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-an-epic-ai-debateand-why-everyone-90c73e-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-an-epic-ai-debateand-why-everyone-90c73e-a1.