“As I emphasized last year in front of the Senate, there is not one risk but many: bias, disinformation, defamation, toxicity, invasion of privacy, and so much more, to say nothing of risks to employment, intellectual property, and the environment, precisely because recent techniques can be applied in so many ways, still poorly understood.”
Our reading, the sentence that was checked:
AI poses not one risk but many, including bias, disinformation, defamation, toxicity, invasion of privacy and so much more, plus risks to employment, intellectual property and the environment, precisely because recent techniques can be applied in so many ways and are still poorly understood.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Why California’s AI Safety Bill should (still) be signed into law - and why that won’t be nearly enough, paragraph 21.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-why-californias-ai-safety-bill-should-cc8872-a1
We recorded 21 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 2024-08-20. 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
- there is not one risk but many
- bias, disinformation, defamation, toxicity, invasion of privacy
- risks to employment, intellectual property, and the environment
- precisely because recent techniques can be applied in so many ways, still poorly understood
Our interpretation
- Assess the substantive risk claim as it stood when posted on 2024-08-20; the reference to Senate testimony is background.
- The ordinary reading concerns credible risks, not a claim that every harm has occurred everywhere or that net employment losses are inevitable.
- Read the causal explanation as saying that broad applicability spreads exposure to different harms and limited understanding makes those harms harder to anticipate and control. It does not identify these as the only causes.
- Recent techniques principally refers to generative AI and foundation models in this context.
Counts as supported if: Research documents plausible mechanisms or observed examples across the listed risk categories and explains how broad reuse and gaps in understanding contribute to their spread or management difficulty.
Counts as contradicted if: Strong evidence shows that the listed categories are not credible AI risks, or that broad applicability and limited understanding do not contribute to the claimed range of risks.
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
Research available before the post identifies multiple distinct risks matching the listed categories. It also explains how broad reuse propagates model defects and how limited understanding complicates anticipating and controlling harm. This supports the ordinary claim about risks and contributing mechanisms. It does not establish that every harm is inevitable or that these are the only causes.
Limitations
- The evidence includes scientific syntheses and a preprint, rather than one experiment testing the entire claim.
- Employment disruption does not imply a proven net loss of jobs.
- Defamation risk does not establish legal liability in any particular case.
- The claim gives no numerical magnitude or exclusive causal model.
Evidence (3 items)
Supports the claim: National Institute of Standards and Technology,
difficult to estimate
Why it matters here: The report identifies bias, harmful content, disinformation, privacy, intellectual-property and environmental risks, and discusses labor-market effects. It links estimation difficulties to diverse applications, limited training-data visibility, and immature safety measurement.
- Publisher
- National Institute of Standards and Technology
- Kind of source
- Secondary summary
- Describes events of
- Public from
Supports the claim: Stanford Center for Research on Foundation Models; arXiv,
adapted models downstream.
Why it matters here: The paper explains a mechanism through which reuse across applications spreads defects. Its abstract also describes limited understanding of model behavior and failures, supporting the causal explanation without establishing an exclusive cause.
- Publisher
- Stanford Center for Research on Foundation Models; arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: UK Department for Science, Innovation and Technology; international scientific expert group,
This could mean many people could lose their current jobs.
Why it matters here: The assessment connects broad task automation to employment disruption and discusses synthetic media used to defame political opponents. It cautions that new jobs could offset losses and that evidence about campaign effects remains limited.
- Publisher
- UK Department for Science, Innovation and Technology; international scientific expert group
- Kind of source
- Secondary summary
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Why California’s AI Safety Bill should (still) be signed into law - and why that won’t be nearly enough
As I emphasized last year in front of the Senate, there is not one risk but many: bias, disinformation, defamation, toxicity, invasion of privacy, and so much more, to say nothing of risks to employment, intellectual property, and the environment, precisely because recent techniques can be applied in so many ways, still poorly understood.
- Where
- paragraph 21
- Published
- 2024-08-20
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-why-californias-ai-safety-bill-should-cc8872-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-why-californias-ai-safety-bill-should-cc8872-a1.