A majority want government to keep AI under human control
The clearest finding in the survey: oversight is not a fringe position.
Oversight
Policy desk · The AP-NORC findings
An AP-NORC poll finds many Americans think AI is moving faster than they can follow, and most say government should make sure the technology stays under human control. From the technology desk at Elisha Yong Journal, an independent news blog covering technology, business, policy, sports, local New York stories and world events.
What the poll asked
Respondents told pollsters the industry is releasing new tools faster than workplaces, schools and households can adjust to them.
Most of those surveyed said the government should make certain AI systems remain under human control rather than being left to decide on their own.
People who said they use AI tools regularly still described themselves as unsure about what those tools do with the information they are given.
Reporting by the Elisha Yong Journal policy desk
The AP-NORC survey lands in a year when AI moved from a topic people read about to something they meet at the pharmacy counter, in their insurance paperwork and on the phone with a bank.
That shift explains the shape of the answers. Concern about the pace of artificial intelligence development is not, in the poll, tied to a single industry or a single company. It reads more like a general unease that the ground rules are still being written while the products are already in use. When a technology shows up inside everyday transactions before the public has a clear picture of it, questions about who is accountable tend to follow.
The poll also measures a quieter expectation: that someone, somewhere, is keeping score. For readers following the technology beat, that is the thread worth pulling — not whether Americans are for or against AI, but what kind of oversight they assume already exists and how much of that assumption is correct.
Four readings from the AP-NORC poll, each with the practical question it raises for anyone who uses these tools at work or at home.
The clearest finding in the survey: oversight is not a fringe position.
Oversight
New models and tools arrive faster than training, policy or plain experience can absorb them.
Pace
Regular users describe the technology differently from people who have barely touched it.
Trust
Concern about AI development is running ahead of any expectation that legislation will catch up this year.
Timing
Read the fuller numbers and the survey’s own framing at the source before quoting them; poll wording and question order both shape what an answer means.
The poll numbers do not sit alone. Cities, regulators and courts have been handling the same questions with far less patience for abstraction.
Policy
Council members spent the session pressing on who carries responsibility when an automated system makes a bad call — the vendor, the operator, or the city agency that bought it.
Read the hearing reportTechnology desk
Business desk
Both stories turn on the same question the poll keeps surfacing: when an automated system makes a decision, who answers for it.
See every sectionWhy the survey is worth your attention
Polling is a crude instrument. It cannot tell you what a careful person thinks after an hour of reading. What it can do is show which worries have moved from the opinion pages into ordinary conversation, and this survey suggests artificial intelligence has made that move.
For anyone hiring, teaching, running a small business or handling claims for a public agency, that matters practically. Public expectations shape what vendors will promise, what regulators will risk, and how much explanation a customer will accept.
Four moods, one survey. None of them is a verdict on whether AI is good; all of them describe how ready people feel.
Polls are one input among many. Before we run a survey story, we check the wording of each question, the sample it was drawn from, and whether the published release explains its own limits. When a number is contested, we say so rather than rounding it into certainty.
The same standard applies across the technology and policy beats, where vendors, agencies and advocacy groups all have something to gain from how a finding is framed.
If something here is unclear or wrong, write to the newsroom and we will look at it. Corrections run at the foot of the story, not buried elsewhere.
ContactThe survey was conducted by AP-NORC, the polling arm that pairs the Associated Press with the research center at the University of Chicago. Its questions covered how respondents view the speed of development, how much control they want humans to keep, and how much they trust the people building these systems.
Not from the answers we reviewed. Wanting oversight is not the same as wanting a ban. Most respondents who described the pace as too fast also said the technology has uses they value, which is why the control question gets more agreement than any question about halting development.
Start with the exact wording of each question and the order it appeared in. A question asked after several others about job losses will draw a different answer than the same question asked cold. Look for the margin of error and the sample size before treating any gap between two groups as meaningful.
Yes, as bills move and city councils take up the questions the poll describes. Those stories land on the policy and technology desks, and our archive stays open so you can trace how a claim changed from one month to the next.
Technology, policy and business all now run into the same questions about automated decision-making. The archive collects every story by desk, and the editorial standards page explains how we source, verify and correct what we publish.
The newsroom
Elisha Yong Journal publishes from 354 E 31st St, Brooklyn, NY 11226. The desk answers weekdays between 9:00 AM and 6:00 PM.
Corrections and reader notes go to the same address; we log them against the story they concern.