Technology & Science
Most Americans say AI is moving too fast, AP-NORC poll finds
The poll found broad concern about AI’s pace, but little confidence in either party’s ability to handle the technology and limited stated use of AI for election information.
Executive summary
- A majority says the pace is too fast: In an AP-NORC survey of 2,140 U.S. adults conducted Sept. 24-28, 64% said artificial intelligence in the United States is developing too fast. Twenty-seven percent said the pace is about right, and 8% said it is not fast enough. The overall margin of sampling error was 2.9 percentage points.[1][2][3]
- Human control and jobs rank highly: Asked to rate government priorities separately, 81% said keeping AI under human control was extremely or very important, and 77% said the same about protecting U.S. workers. The survey did not ask respondents to choose one priority over another.[2][1]
- No party has a clear trust advantage: Thirty-nine percent said neither party could be trusted to handle AI well, compared with 23% who chose Democrats and 21% who chose Republicans. A separate Reuters/Ipsos poll, using different questions and a registered-voter subsample, also found concern about AI and jobs.[2][3][4]
Nearly two-thirds of U.S. adults surveyed by the Associated Press-NORC Center for Public Affairs Research said AI technology in the country is developing too fast. The September poll put that share at 64%, against 27% who said the pace was about right and 8% who said it was not fast enough. The question asked about the pace of development, not whether respondents wanted a particular law, ban or pause.[1][2][3]
NORC conducted the survey Sept. 24-28 with 2,140 adults through its AmeriSpeak panel. The overall margin of sampling error was plus or minus 2.9 percentage points. AP-NORC says the probability-based panel is designed to represent U.S. adults; subgroup estimates have wider margins, including 4.3 points for Democrats and 4.7 for Republicans.[2][5]
When asked how important different goals should be in the U.S. government’s approach to AI, 81% rated keeping the technology under human control extremely or very important. Seventy-seven percent gave that rating to protecting U.S. workers, 69% to minimizing environmental harm and 63% to growing the economy. Promoting innovation received that rating from 51%, while 45% said it was important to compete with countries such as China.[2][1]
These were separate ratings, not a forced ranking or a trade-off question. The results show which outcomes respondents say matter, but do not establish how they would balance those goals in a specific policy or what rules they support. The poll’s findings therefore describe public priorities rather than a mandate for one regulatory approach.[2][1]
Asked which party they trusted to do a better job handling AI, 39% said neither. Twenty-three percent chose Democrats, 21% chose Republicans and 16% said both parties equally. The poll also found that 31% approved of President Donald Trump’s handling of AI, while 67% disapproved.[2][1][3]
The pace question showed a partisan difference without erasing concern in either group: AP-NORC reported that 76% of Democrats and 54% of Republicans said AI was developing too fast. The survey does not explain why respondents hold those views or whether they will affect voting or policy decisions.[3][2]
The survey also asked about using AI in connection with the upcoming election. Twenty-one percent said they had used or planned to use it to find information about a candidate or ballot issue. Twelve percent said the same about finding out how to vote, and 10% about how to register or deciding whom to vote for. Those figures combine reported past use and stated plans; they are not measurements of completed use alone.[2][1]
A Reuters/Ipsos poll published Oct. 7 found that 84% of voters viewed AI as a threat to U.S. workers and that 80% of Democrats and 61% of Republicans supported stricter government regulation. That survey included 4,506 U.S. adults, 3,526 of them registered voters. It asked different questions from AP-NORC and used a different population, so its results cannot be read as a direct comparison or a change over time.[4]