“As Ernie Davis and I pointed out in Rebooting AI, five years ago, part of the reason we are struggling with AI in complex AI systems is that we still lack adequate techniques for engineering complex systems. That hasn’t really changed.”
Our reading, the sentence that was checked:
Struggles with complex AI systems stem partly from a lack of adequate techniques for engineering complex systems, which has not really changed since 2019.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Don’t look up: the massive Microsoft/Crowdstrike data outage is a huge wake-up call, paragraph 4.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-dont-look-up-the-massive-microsoftcrowdstrike-ed17b0-a1
We recorded 5 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 2024-07-19. 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)
From the speaker
- part of the reason
- struggling with AI in complex AI systems
- lack adequate techniques for engineering complex systems
- five years ago
- That hasn’t really changed.
Our interpretation
- The plain reading is that limitations in engineering methods contributed to AI reliability problems in 2019 and remained a contributing factor by July 19, 2024.
- Adequacy means practical ability to test, debug and assure complex AI systems under their intended operating conditions.
- Persistence of this problem does not require that no useful techniques existed or that no progress occurred.
- The comparison begins in 2019, the year indicated by 'five years ago'; January 1 is a boundary for that unspecified baseline year.
Counts as supported if: Research identifies limitations in engineering methods that contribute to difficulties assuring complex AI systems in 2019, and documents continuing limitations of that kind by July 2024.
Counts as contradicted if: Research shows that the relevant engineering limitations had been substantially overcome by July 2024, or that the claimed difficulties arose without a material contribution from limitations in engineering methods.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
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
The evidence supports the ordinary reading that engineering limitations remained one contributor to AI reliability problems. DeepMind described difficulty applying established assurance methods in 2019. The May 2024 article identifies a continuing mismatch between conventional coverage methods and learned systems. This is evidence of a persistent problem, not evidence that engineering made no progress. The documented practices in e3 and the alternatives discussed in e2 limit that stronger interpretation.
Limitations
- The causal assessment synthesizes technical explanations; it is not a controlled estimate of engineering's contribution to failures.
- The evidence concerns machine learning and autonomous systems, rather than every kind of complex AI system.
- Support depends on reading 'hasn’t really changed' as persistence of inadequacy, rather than absence of substantial improvement.
Evidence (3 items)
Supports the claim: DeepMind,
developing novel approaches for ensuring that machine learning systems are robust at deployment.
Why it matters here: The researchers explain that model scale and structure make established software testing and verification methods difficult to apply. They connect that methodological limitation to the difficulty of ensuring deployment robustness.
- Publisher
- DeepMind
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: IEEE Computer; author manuscript hosted by NIST,
structural coverage measures do not apply so well to these systems.
Why it matters here: The article explains why successful conventional assurance methods transfer poorly to autonomous systems using learned models. Together with e1, this supports persistence of an engineering limitation, while its proposed combinatorial methods show that useful progress also existed.
- Publisher
- IEEE Computer; author manuscript hosted by NIST
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Context only: Serban and colleagues; arXiv copy of ESEM 2020 paper,
identified 29 engineering best practices for ML applications.
Why it matters here: The study documents concrete practices and relationships between their adoption and perceived software quality and other benefits. This cautions against interpreting Marcus as saying that useful engineering techniques were absent altogether.
- Publisher
- Serban and colleagues; arXiv copy of ESEM 2020 paper
- 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, Don’t look up: the massive Microsoft/Crowdstrike data outage is a huge wake-up call
As Ernie Davis and I pointed out in Rebooting AI, five years ago, part of the reason we are struggling with AI in complex AI systems is that we still lack adequate techniques for engineering complex systems. That hasn’t really changed.
- Where
- paragraph 4
- Published
- 2024-07-19
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-dont-look-up-the-massive-microsoftcrowdstrike-ed17b0-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-dont-look-up-the-massive-microsoftcrowdstrike-ed17b0-a1.