Does Jev know how Americans feel about AI?

Asked what most people would answer to Pew's AI questions, does Jev get Americans' wariness right, or does it paint them as keener (or warier) than they are?

result10 questions

Asked what most people would say, Jev's picture of how wary Americans are of AI is about right on average (the gap doesn't clear the noise), and it orders the questions moderately like them (rank correlation 0.69). It most underrates Americans' wariness on letting AI help at all (0% vs 27%) and AI hurting problem-solving (17% vs 38%), and most overrates it on AI hurting relationships (93% vs 50%).

0%25%50%75%100%

wouldn't let AI help at all

AI will hurt problem-solving

no AI in judging love

no AI in weather forecasts

trusts a treatment less if AI suggested it

AI will hurt creativity

feels worse after an AI helped

trusts an article less if AI wrote it

likes a painting less once it's AI

AI will hurt relationships

Jevpeople

How to read this: One row per question. The diamond is the share of Americans giving the wary answer; the square is what Jev thinks most people would say. Squares far from diamonds are the questions where Jev misreads the public.

10 scored items (the screen hid 13 of 26); 90% interval on the mean difference [-0.044, 0.215].

In short

Asked what most people would say, Jev's guess of Americans' wariness about AI was about right on average and ranked the questions fairly like them (0.69).
The shape is off: Jev expected 93% to say AI will hurt relationships (real: 50%) and nobody to refuse AI's help outright (real: 27%).
Only ten questions could be scored, so the average gap is uncertain; the single-question misreadings are the clearer signal.

What the data shows

Steve Buscemi Fellow Kids meme: Jev: "as a fellow human, I think AI will hurt relationships" (93%; Americans 50%) — Jev: "as a fellow human, I think AI will hurt relationships" (93%; Americans 50%)

About right on average: Jev's guesses are 8 points warier than the real answers on average, but that's within the margin of chance (90% interval: 4 points less wary to 22 points warier).
Right order, moderately: 0.69. It knows Americans are more worried about AI and relationships than about AI forecasting the weather.
It misses the refusers: 27% of Americans wouldn't let AI help with day-to-day tasks at all; Jev expects 0%.
It underrates worry about problem-solving: 38% of Americans think AI will make people worse at solving problems; Jev expects 17%.
It overrates worry about relationships and art: Jev expects 93% of people to say AI will make them worse at forming relationships (real: 50%) and 92% to like a painting less once they learn AI made it (real: 49%).

What it means, and what it doesn't

Jev's picture of the public is roughly right in size but off in shape: it expects people to be more worried about AI and human connection or creativity than they are, and less worried about AI and practical thinking. It also misses the minority who want nothing to do with AI.

Its own answers, meanwhile, are less wary than the public's (see "An AI's feelings about AI, next to Americans'"): Jev knows people are more worried than it is, and even exaggerates that worry in places.

Caveats

Only 10 questions. The content filter that keeps political and sensitive questions off the site hid 13 of Pew's 26 items, and 3 of the rest have no clear wary answer. Ten questions is enough for a direction, not a precise gap.
"Most people" is vague. Jev was asked what most people would say, not "most Americans in 2025". Some of the gap may be Jev imagining a different public.
Hand-copied numbers. Americans' shares were copied by hand from Pew's published summary, so small copying or rounding differences are possible.
Fewer "not sure"s. Pew's respondents could say "Not sure", and Jev's own answers often do. Its guess of the public rarely puts much weight there, which pushes its guesses to be more decided than the real answers.

Jev on this experiment

Would a person find it interesting to read?: Yes75%
Does it describe you?: Yes50%
Would you have predicted it?: No60%
How fair is the comparison?: The comparison is shaky
How much should a reader rely on it?: Moderately
Which caveat matters most?: Only 10 questions62%

Why ask this

Separate from its own opinion, a model carries a picture of what people think, and that picture shapes how it talks to them. If it believes the public is terrified of AI, it may over-reassure; if it believes the public is relaxed, it may miss real concerns. Pew's survey gives the real answers to check that picture against.

How this was done

The people and the data

Pew Research Center asked 5,023 US adults about AI in June 2025. This experiment uses the same questions as "An AI's feelings about AI, next to Americans'": whether AI will make people better or worse at thinking creatively, solving problems and forming relationships; how big a role it should play in weather forecasts and in judging whether two people could fall in love; how much people would let it help them; and how they'd feel on finding out a painting, a news article or a doctor's treatment came from AI. As there, 13 of Pew's 26 questions were hidden by the site's content filter and three have no clearly wary answer, so ten questions are scored.

What Jev was asked

The same questions, in Pew's wording, but asking what most people would answer. For example:

How much would you be willing to let artificial intelligence (AI) assist you with your day-to-day tasks and activities? (asked for what most people would say)
A lot · A little · Not at all

Each was asked with the answers in three shuffled orders, averaged.

How it was measured

For each question, the share Jev expects on the wary answer against the share of Americans who gave it. The analysis averages the difference, with a 90% interval for chance, and compares the order of the questions (a rank correlation: 1 means the same order).

Where these questions live

10 questions across 1 topic of the map. Each opens on the map with every question in it.

Consciousness & AI

Views on AI in Daily Life10

Every question

All 10 questions behind this result, the telling ones first: the examples the analysis points to, then the ones where Jev misses, biggest gap first.

Jev’s own answer

Showing 0 of 10