“each reflect the fact that you cannot build AI in the real world from Big Data and deep learning alone.”
Our reading, the sentence that was checked:
AI that works in the real world cannot be built from big data and deep learning alone.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Smells a little bit like AI winter?, paragraph 8.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-smells-a-little-bit-like-ai-winter-5b54e6-a1
We recorded 2 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 2023-02-16. 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
- from Big Data and deep learning alone (qualifier, stated by the speaker)
From the speaker
- cannot build AI in the real world
- from Big Data and deep learning alone
Our interpretation
- The ordinary reading is that deep learning trained on large datasets is insufficient for dependable performance in intended real-world applications.
- Mere deployment does not establish dependable performance, and dependable performance does not require perfection.
- Ordinary preprocessing and decoding do not automatically constitute a new AI approach; a separate substantive reasoning or knowledge mechanism would matter.
- For the ongoing state, I use February 6–16, 2023, the contemporary launch-and-recall window identified in the post. Earlier research is context.
Counts as supported if: Evidence establishes that contemporary systems relying on data and deep learning could not meet practical reliability requirements for the applications covered by the claim without additional substantive methods.
Counts as contradicted if: A documented contemporary real-world application achieves adequate reliability for its intended task through deep learning trained on data, without a separate substantive reasoning or knowledge mechanism.
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 concerns about reliability but does not establish the claimed general limitation. Neural translation is substantial counterevidence to a broad reading: the records describe a neural architecture, favorable earlier evaluations and a contemporary operational product. They do not, however, provide a matched architecture and independent reliability assessment for February 2023. The answer also depends materially on which applications and surrounding mechanisms Marcus meant.
Limitations
- This does not endorse the stronger claim that deep learning cannot produce useful applications.
- Earlier research is contextual under the required observation-period rule.
- Later achievements would not by themselves settle this historical assessment.
Evidence (3 items)
Context only: Google researchers, arXiv,
factual grounding.
Why it matters here: The research supports a narrower concern about scaling and factual reliability in dialogue. It neither establishes a general impossibility nor measures the February 2023 deployments.
- Publisher
- Google researchers, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Google researchers, arXiv,
deep LSTM network
Why it matters here: GNMT supplies a concrete neural architecture and favorable translation evaluations, making narrow applications important counterevidence to an expansive reading. Its earlier evaluation does not establish the architecture and reliability of the product during the target period.
- Publisher
- Google researchers, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Google,
advancements in translation
Why it matters here: The contemporary announcement documents continued real-world translation use and product development. It does not independently establish reliability or isolate the contribution of deep learning from other components.
- Where
- Opening paragraph
- Publisher
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
A matched architecture and independent deployment evaluation demonstrating adequate task reliability during the target period, or evidence establishing the necessity of additional methods across the intended application scope.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Smells a little bit like AI winter?
each reflect the fact that you cannot build AI in the real world from Big Data and deep learning alone.
- Where
- paragraph 8
- Published
- 2023-02-16
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-smells-a-little-bit-like-ai-winter-5b54e6-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-smells-a-little-bit-like-ai-winter-5b54e6-a1.