What was published
On 11 November 2026 Public Organization Review published an open-access study by Yeonkyung Kim, Yi-Fan Wang and Grace Jung. They recruited 3,010 U.S. respondents through an online platform and asked about their attitudes toward AI in public services. Each respondent then read two short scenarios, a tax return and a food-assistance (SNAP) application, in which either an AI system or a human manager handled the case with a good or a bad outcome, and then answered the questions again.
As the authors report it, those who read about AI came to believe more strongly that AI can outperform human managers, and yet their overall attitude toward AI fell compared with those who read about human managers. Negative AI outcomes lowered trust more than positive ones raised it, most strongly in the SNAP scenario. The authors call AI “competent but unwelcome” and conclude that efficiency alone does not generate legitimacy.
What we notice
The study asks how people feel about AI. We found ourselves asking what the person in the scenario is given. In the negative SNAP versions, the applicant is told to start over, with the problems found pointed out. The scenario does not give a reason addressed to the applicant, a named person to ask, or a way to say that their situation has been misread. That is how we read the appendix; the study did not vary these things, so it cannot tell us whether they matter.
The authors suggest human oversight and transparency as ways to bridge the gap. That is their suggestion. Our own reading is only a question: perhaps part of what people withhold from a machine is the sense of being addressed by someone. The study does not say so.
A proposed illustration
This is a hypothetical, not a finding of the study. A man who has just lost his job applies online for food assistance and is told only that the application is incomplete. Whether a system or a clerk produced that sentence, he is left with the same questions: what is wrong, whom can I ask, will anyone read what I say? At a counter, a clerk can be asked. A screen may not offer the same thing.
What the study does not say
It does not involve real claimants. The authors acknowledge that vignettes cannot replicate the stakes of real welfare denials and that an online sample may leave out people who are digitally vulnerable. The scenarios always came in the same order, tax then SNAP, so the second answer may carry traces of the first. It does not ask about dignity, being heard or being understood, and the authors’ own measures are attitudes and trust.
If competence is not what people are missing, what are they asking an institution for when they come to it in need?
Signed articles reflect the views of their authors. Corrections: info@dghumana.com.
