“With technology more and more able to emulate human responses based on larger and larger samples of text (and “reinforcement learning” from humans who instruct the machines), the problem has grown even more pernicious.”
Our reading, the sentence that was checked:
Growing ability of AI to emulate human responses has made the overattribution (ELIZA effect) problem more pernicious.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Stop Treating AI Models Like People, paragraph 6.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-stop-treating-ai-models-like-people-a5113a-a1
We recorded 9 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 2023-04-17. 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
- more and more able to emulate human responses
- larger and larger samples of text
- “reinforcement learning” from humans
- the problem has grown even more pernicious
Our interpretation
- The problem refers to mistaken attribution of human qualities and the harmful reliance discussed immediately afterward.
- More pernicious ordinarily means that improved humanlike dialogue increases the risk or severity of such mistaken attribution and harmful reliance.
- The historical comparison runs from the earlier ELIZA-style systems discussed in the context to systems available by April 17, 2023. No numerical increase or exact baseline year is supplied.
- A comparison of perceived warmth or satisfaction alone would not establish increased harmful reliance.
Counts as supported if: Comparative causal evidence shows that improved humanlike dialogue increases mistaken attribution and consequential harmful reliance, with a defensible connection to the historical change claimed.
Counts as contradicted if: Relevant controlled comparisons show that greater humanlike capability does not increase, or reduces, mistaken attribution and harmful reliance across the settings covered by the claim.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
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
Comparative research supports a plausible mechanism: beliefs about an AI's motives can matter more with a more sophisticated conversational system. That does not establish that harmful overattribution had increased by the post date. The available evidence does not measure the claimed historical change in harm or isolate the training changes named in the sentence.
Limitations
- The central comparative paper appeared after the post, although before the evaluation cutoff.
- Its reported effects concern perceptions, not verified harmful outcomes.
- Several potentially relevant full texts could not be retrieved.
Evidence (1 item)
Context only: Nature Machine Intelligence; author manuscript hosted by MIT,
more sophisticated conversational agents
Why it matters here: The October 2023 report compares GPT-3 and ELIZA and finds stronger effects of perceived motives with the more sophisticated system. It supports a possible mechanism, but does not establish increased harmful reliance by April 2023; the experiment's collection date was not established.
- Publisher
- Nature Machine Intelligence; author manuscript hosted by MIT
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
What would settle it
Comparable studies of earlier and April 2023 systems that measure mistaken attribution and harmful reliance, control competing explanations, and establish the timing of the change.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Stop Treating AI Models Like People
With technology more and more able to emulate human responses based on larger and larger samples of text (and “reinforcement learning” from humans who instruct the machines), the problem has grown even more pernicious.
- Where
- paragraph 6
- Published
- 2023-04-17
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-stop-treating-ai-models-like-people-a5113a-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-stop-treating-ai-models-like-people-a5113a-a1.