“The AI world pretty much divides into two groups: those who understand why current machine learning techniques suck at outliers, and therefore struggle at things like driverless cars and high-level reasoning in unusual circumstances — and those who don’t.”
Our reading, the sentence that was checked:
Current machine learning techniques handle outliers poorly and so struggle with driverless cars and high-level reasoning in unusual circumstances.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, This one important fact about current AI explains almost everything, paragraph 17.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-this-one-important-fact-about-current-f263c7-a1
We recorded 10 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 2024-08-01. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- 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
- in unusual circumstances (qualifier, stated by the speaker)
From the speaker
- current machine learning techniques suck at outliers
- and therefore struggle
- driverless cars
- high-level reasoning in unusual circumstances
Our interpretation
- The assessable proposition is the causal explanation identified in the supplied normalization. The rhetorical division of the AI community is not treated as a separate population claim.
- The plain reading is that poor handling of unusual inputs contributes to failures in both automated driving and reasoning. It need not be the sole cause.
- A failure in an unusual situation is insufficient by itself: evidence must connect that failure to the system's handling of the unusual input.
- I use 2024 through publication for the contemporary systems described. Earlier crashes can provide context but cannot establish a defect still present in that period.
Counts as supported if: Controlled reasoning evaluations and driving investigations connect unusual-input handling to failures in both domains, rather than merely documenting errors or crashes.
Counts as contradicted if: Controlled evidence shows that the relevant failures persist independently of unusual inputs and are explained by other mechanisms, while the systems reliably handle the alleged outliers.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
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 reasoning experiments support a contribution from sensitivity to altered inputs. The driving filing documents a relevant failure and its software conditions, but does not establish a failure to generalize beyond training examples. Nor do the reasoning experiments fully separate unfamiliarity from increased difficulty. The evidence is consistent with Marcus's explanation, but does not settle the causal claim across both domains. An unproven component is not counterevidence, so a mixed verdict would be inappropriate.
Limitations
- The internal training data and model-level causes of the driving failure are unavailable in the cited filing.
- This finding does not establish that outlier handling was irrelevant to the collision.
- The cited reasoning study uses selected perturbations rather than a representative sample of all reasoning tasks.
Evidence (2 items)
Supports the claim: arXiv; Hong and colleagues,
perturbed questions significantly challenge all models
Why it matters here: Controlled question alterations produced substantial performance declines, supporting the reasoning component. Section 6.1 acknowledges that some alterations increase reasoning difficulty, so the experiment does not isolate unfamiliarity as the sole mechanism.
- Publisher
- arXiv; Hong and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: NHTSA; submitted by Waymo LLC,
increased risk
Why it matters here: The filing connects a pole collision to a combination of mapping and perception conditions, including a low predicted damage score. It does not establish that the object was outside training coverage or isolate a machine-learning generalization failure from other software mechanisms.
- Publisher
- NHTSA; submitted by Waymo LLC
- Kind of source
- Regulatory filing
- Describes events of
- Public from
What would settle it
A driving investigation or controlled replay that isolates poor handling of unusual inputs as a cause, together with reasoning tests that hold task difficulty constant and establish the same causal connection.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, This one important fact about current AI explains almost everything
The AI world pretty much divides into two groups: those who understand why current machine learning techniques suck at outliers, and therefore struggle at things like driverless cars and high-level reasoning in unusual circumstances — and those who don’t.
- Where
- paragraph 17
- Published
- 2024-08-01
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-this-one-important-fact-about-current-f263c7-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-this-one-important-fact-about-current-f263c7-a1.