A clear majority of Americans — 57% — now say artificial intelligence is bad for society, and 67% disapprove of how President Donald Trump is handling the technology. [1] Those two figures, from an ABC News/Washington Post/Ipsos poll released Wednesday, have been read mostly as a verdict on the White House. They are also something subtler: a map of where the public believes the technology genuinely helps, and where it quietly fails to.
The Gain People Report for Themselves
The most telling number in the survey is not the headline disapproval. It is the gap between the 42% who consider AI at least somewhat good for society and the half who call it at least somewhat good for them personally. That gap of roughly eight points is the whole story in miniature. People are finding something useful in the tools they touch directly, while remaining uneasy about what those same tools do at scale. The personal benefit is real enough to register; the collective consequence is worrying enough to offset it. The poll measures expectations, not outcomes, but a respondent who says AI helps them is describing a gain they believe they have already felt.
What the Skepticism Is Actually About
It is worth being precise about what the negative responses target. More than six in 10 respondents see a high or very high risk that AI will eliminate many jobs. Similar majorities worry about the technology penetrating secure computer systems and about it weakening people’s ability to make decisions for themselves. Fewer than a third see a big chance that AI creates many jobs, simplifies daily tasks, or delivers technological dominance. Read together, this suggests these are not objections to capability. They are objections to distribution and control — who captures the gain, and who decides how it is used. The concern sits downstream of the improvement, not against it.
The Oversight Standard
Roughly two-thirds of Americans say the federal government is doing too little to regulate AI’s growth and development, and a majority of Republicans share that view. [1] This is the question of the standard — who defines what good enough looks like, and on what evidence. The same survey shows about two-thirds oppose building new data centers nationwide, with roughly half of Republicans and 76% of Democrats opposed, and majorities in both parties rejecting construction near their own homes. [1]. The demand crosses party lines, which means it is not simply a proxy for opposition to the president. It is a request for a benchmark of acceptable performance before the technology is allowed to spread further. Neither party currently owns that trust: nearly half of Americans say they trust neither Democrats nor Republicans to handle AI, with 26% backing Democrats and 19% backing Republicans. [1]
Where the Gains Are and Are Not
The survey does not claim AI is failing to improve anything, and it measures expectations rather than outcomes. It shows that the improvements people can name are personal and practical, while the improvements they cannot yet see are the ones that would justify the buildout. That is the gap the next round of policy has to close. A technology that half the country says helps them individually has a foundation to build on. Whether it earns broader assent depends on whether the gains become visible beyond the individual user — in jobs created rather than displaced, in systems made more secure rather than more exposed, and in decisions people feel they still control. The poll does not settle that question. It tells us where the argument now stands.
