Questions and method

These charts use unweighted descriptive results from two McQueen Analytics surveys conducted in May and June 2026. They describe the respondents in each source. They are not population estimates, a time series or tests of a frontier slowdown.

Human review and current AI use

Survey 2: May 24–June 12, 2026. Base: 2,584. All selected answers were present and recognized; no new exclusions or weighting were applied.

QID51: Which comes closest to how you use AI tools today?

Response choices: I have not used AI tools.; I tried one once or twice.; I use AI for simple questions or quick answers.; I use AI to help write, summarize, plan, or brainstorm.; I go back and forth with AI and check its work.; I use AI as part of a regular workflow with files, data, code, images, or other work.; Not sure..

QID80: How important would each of these be for trusting AI use in a workplace or school? Workers or students can ask for a human review of important decisions.

Response choices: Not at all important; A little important; Very important; Extremely important.

Human-review percentages combine Very important and Extremely important. Labels in the graphic abbreviate the exact use responses below. The two smallest groups are flagged in the chart. No causal effect or hypothesis test is claimed.

Exact AI-use responseBaseVery / extremely importantShare
I have not used AI tools.1297457.4%
I tried one once or twice.31621367.4%
I use AI for simple questions or quick answers.75358577.7%
I use AI to help write, summarize, plan, or brainstorm.54745382.8%
I go back and forth with AI and check its work.17814279.8%
I use AI as part of a regular workflow with files, data, code, images, or other work.60456393.2%
Not sure.572543.9%

How much of a role should AI play?

Survey 4: May 24–June 8, 2026. Each item has a base of 1,201. All selected answers were present and recognized. The illustrated charts show all four choices; the classic charts combine the two supporting roles. Decision-making does not mean unlimited or wholly unsupervised autonomy.

  1. No role at all
  2. A small supporting role
  3. A major supporting role, with a person responsible for the final decision
  4. A major role that can make decisions

QID111

How much of a role should AI play in each of the following? Helping doctors review a diagnosis or treatment plan.

ResponseCountShare
No role at all20116.7%
A small supporting role47539.6%
A major supporting role, with a person responsible for the final decision38732.2%
A major role that can make decisions13811.5%

QID112

How much of a role should AI play in each of the following? Helping perform parts of surgery.

ResponseCountShare
No role at all30525.4%
A small supporting role36830.6%
A major supporting role, with a person responsible for the final decision38732.2%
A major role that can make decisions14111.7%

QID120

How much of a role should AI play in each of the following? Watching public places for safety risks.

ResponseCountShare
No role at all14812.3%
A small supporting role40433.6%
A major supporting role, with a person responsible for the final decision44136.7%
A major role that can make decisions20817.3%

QID121

How much of a role should AI play in each of the following? Helping parents or schools filter unsafe online content for children.

ResponseCountShare
No role at all13411.2%
A small supporting role37030.8%
A major supporting role, with a person responsible for the final decision47539.6%
A major role that can make decisions22218.5%

Reading the figures

Every quantitative bar starts at zero. Full tracks represent 100% within a use group or survey item, not equal numbers of respondents across use groups. Lengths use exact counts divided by bases; displayed labels round to one decimal. Totals can round to 99.9% or 100.1%. Drawings identify settings or concepts and do not encode quantities.

The classic charts show 95% Wilson binomial intervals. These assume a binomial model; they do not correct selection bias or supply a national margin of error. The illustrated versions omit intervals for readability; counts and bases remain visible here. Preferences for oversight do not establish technical safety.

Research before this debate

The June 23 research directions included acceptance with conditions, experienced users’ interest in safeguards, and human choice and recourse. They preceded September’s executive statements. We did not predict those announcements or test the effectiveness of their proposals.