“Seven years of AI has taught us that deep learning is unpredictable, and not always human like.”
Our reading, the sentence that was checked:
The past seven years of AI have shown that deep learning is unpredictable and not always human-like.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, David beats Go-liath, paragraph 9.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-david-beats-go-liath-2137d6-a1
We recorded 4 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 2023-02-18. 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
- Seven years of AI
- deep learning is unpredictable
- not always human like
Our interpretation
- The surrounding reference to March 2016 anchors the approximately seven-year period.
- The ordinary reading is that successful deep-learning systems have exhibited surprising failures and sometimes behaved differently from humans on comparable tasks.
- Unpredictable is read as unexpected failure under changed inputs, not mathematical randomness or an inability to predict any output.
Counts as supported if: Research during the stated period documents surprising failures and observable differences between deep-learning systems and humans.
Counts as contradicted if: The relevant comparisons show that the alleged failures were not reproducible or that the purported behavioral differences disappear under appropriate controls.
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
Research within the period demonstrates unexpected failures and clear human–model behavioral differences. That supports the ordinary, limited reading of 'unpredictable' and 'not always human like.' Other experiments show that humans can sometimes anticipate machine errors, so a claim of complete unpredictability would go beyond the evidence.
Limitations
- The result depends on reading unpredictable as surprising failures rather than universal unpredictability.
- These studies do not establish a universal difference between human cognition and every deep-learning architecture.
Evidence (3 items)
Supports the claim: Geirhos et al., ICLR paper hosted on arXiv,
CNN recognition accuracies were much lower than human accuracies.
Why it matters here: Humans and networks classified the same images differently. The experiments also identified differing reliance on shape and texture.
- Publisher
- Geirhos et al., ICLR paper hosted on arXiv
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Supports the claim: labsix,
classifies as an espresso at every angle!
Why it matters here: The authors demonstrated a physical object that produced a conspicuously unexpected classification across viewpoints. This supports the ordinary claim about surprising failures.
- Publisher
- labsix
- Kind of source
- Demonstration
- Describes events of
- Public from
Context only: Nature Communications,
Human intuition may be a surprisingly reliable guide to machine (mis)classification
Why it matters here: Across adversarial-image experiments, people could often anticipate machine labels. This limits any stronger claim that deep-learning behavior is wholly unintelligible or never human-like.
- Publisher
- Nature Communications
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, David beats Go-liath
Seven years of AI has taught us that deep learning is unpredictable, and not always human like.
- Where
- paragraph 9
- Published
- 2023-02-18
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-david-beats-go-liath-2137d6-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.