Rutgers-Led Survey Finds Low Support for Unsupervised AI Decisions
Americans report using AI more often, but the new national survey finds a sharp boundary around systems making consequential choices without a person’s final review.
Listen to this story
The audio brief
Story brief
3 key pointsA Rutgers-led survey of 5,154 U.S. adults found that AI adoption is expanding without equivalent trust in autonomous high-stakes decisions. Usage of major AI tools reached 72%, up from 50% in November 2024, but fewer than 10% support unsupervised AI decisions in areas such as hiring, lending, admissions, parole, or medical prioritization. The findings point toward demand for assistive systems, human oversight, disclosure, and regulation rather than fully automated authority.
- 01
59% favored strict AI regulation, while 25% preferred minimal regulation to encourage innovation.
- 02
Disclosure support was high: 76% for employment and customer decisions, and 77% for student evaluations.
- 03
Daily AI users showed high confidence but limited knowledge: 82% said they understood AI basics, averaging 49% correct on 12 questions.
Americans are bringing AI into daily life far faster than they are willing to hand it authority over their lives. A Rutgers-led national survey found that 72% of U.S. adults have used a major AI tool, yet fewer than one in 10 would let AI make an unsupervised final decision on hiring, lending, college admissions, parole, or priority for a medical procedure.
The result is not a simple rejection of the technology. Twenty-seven percent of adults said they use AI every day. Among employed adults, 44% reported using it at work and 31% said their jobs require it. Reported use of major AI tools rose from 50% in November 2024 to 72% in the latest survey.
Help is more acceptable than final authority
The questionnaire distinguished assistance from AI acting alone. Thirty-four percent of respondents would allow AI to assist with hiring decisions without making the final call, while 26% would allow it to assist with parole decisions. Between 45% and 54% said AI should not be used at all in the surveyed consequential decisions.
Katherine Ognyanova, a Rutgers professor and report co-author, framed the gap as a difference between using an AI tool and trusting it fully. Half of respondents described themselves as worried about AI; that included 48% of daily users. The share who described themselves as angry rose from 19% in late 2024 to 30% in August 2026.
Where respondents want notice
- 76% said employers should disclose AI evaluations of workers or applicants.
- 76% wanted companies to disclose AI decisions about customers.
- 77% wanted schools to disclose AI evaluations of students.
Confidence does not match the knowledge test
The survey also found a knowledge gap that may shape those judgments. Fifty-nine percent said they understood the basics of how AI works, but respondents averaged 48% correct across 12 knowledge statements. Among daily users, 82% said they understood the basics while averaging 49% correct.
Views on whether AI acts in the public interest also shifted by party. In late 2024, 56% of Democrats expressed confidence, compared with 45% of Republicans; by August 2026, the figures were 44% and 56%, respectively. Overall confidence stayed at 46%, while majorities in both parties still favored strict regulation.
The National AI Opinion Monitor surveyed 5,154 U.S. adults online between July 5 and August 1, 2026. Researchers used a nonprobability sample with demographic quotas, then reweighted results across demographic and geographic characteristics. The study measures reported attitudes, not whether particular AI systems can safely or fairly make these decisions.
Editorial analysis
Our Read
The survey points to a practical distinction for organizations deploying AI: public acceptance may depend less on whether a system participates in a decision than on whether it has final authority. That is not evidence that human review makes any particular system fair or safe. It does show that respondents draw a clear line between assistance and autonomous judgment, while also favoring disclosure. The next meaningful evidence will be whether employers, schools, and companies make notice and human review standard parts of consequential AI workflows.
Citation desk / original work
Cite this
Citation desk / original work
Cite this
The survey points to a practical distinction for organizations deploying AI: public acceptance may depend less on whether a system participates in a decision than on whether it has final authority.
/posts/rutgers-led-survey-finds-low-support-for-unsupervised-ai-decisions#finding-1
Sources
- techxplore.comAmericans use artificial intelligence, but still prefer that humans make consequential decisions
Reader comments
Newest comments first. Replies stay oldest first.