“To me the only solution to the long-term risk issue is to build machines with consensus human values, but we are a long way from knowing how to do that”
Our reading, the sentence that was checked:
The field is a long way from knowing how to build machines with consensus human values.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Should we worry more about short-term AI risk or long-term AI risk?, paragraph 9.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-should-we-worry-more-about-short-ec324a-2-a1
We recorded 11 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 2023-03-14. 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
- we are a long way from knowing how
- build machines with consensus human values
- long-term risk
Our interpretation
- The supplied normalized proposition isolates the state of technical knowledge. The separate assertion that this is the only solution is not graded here.
- The ordinary reading of a long way is that substantial foundational problems remained, rather than a forecast of a particular number of years.
- Consensus human values means broadly shared human values reliably reflected in machine behavior across relevant situations.
- For this ongoing research-state observation, the target is the publication month through March 14, 2023. Research has no standard reporting quarter.
Counts as supported if: Contemporaneous research documents substantial unresolved problems in reliably aligning powerful systems with human values, with no established validated general solution.
Counts as contradicted if: An established and validated method available at publication reliably implements broadly shared human values across relevant situations, leaving only routine implementation work.
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
The ordinary technical reading is supported. A contemporaneous primary source explicitly says robust alignment of powerful systems was unsolved and describes foundational research still needed. OpenAI reported meaningful safety improvements, but those did not establish reliable alignment with broadly shared human values. This supports substantial remaining difficulty, not a numerical estimate of how many years a solution would take.
Limitations
- The evidence includes research organizations' own assessments rather than a field-wide proof.
- Consensus values and distance from a solution lack agreed measures.
- The only-solution assertion is outside the supplied normalized proposition.
- The March 14 disclosure may have appeared after the post within that day; the decisive March 8 source predates it.
Evidence (2 items)
Supports the claim: Anthropic,
robustly helpful, honest, and harmless.
Why it matters here: Anthropic explicitly describes robust alignment of powerful systems as unsolved. Its discussion of supervision, generalization and alignment evaluation supports the reading that foundational work remained.
- Publisher
- Anthropic
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: OpenAI,
significantly improved
Why it matters here: This is material evidence of progress, which rules out reading Marcus as claiming that nothing useful was known. The disclosure also describes remaining jailbreaks; improvement against company policies does not demonstrate reliable implementation of consensus values.
- Where
- Risks & mitigations
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Should we worry more about short-term AI risk or long-term AI risk?
To me the only solution to the long-term risk issue is to build machines with consensus human values, but we are a long way from knowing how to do that
- Where
- paragraph 9
- Published
- 2023-03-14
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-should-we-worry-more-about-short-ec324a-2-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-should-we-worry-more-about-short-ec324a-2-a1.