What the survey measured
A Pew Research Center survey of 3,488 US adults, conducted June 22–28, 2026, examined attitudes toward AI in healthcare. Its October report finds that acceptance depends strongly on the task and whether a person reviews the result.
For translating, 26% were open to AI operating alone and 52% to AI with final human review. For diagnosis, the corresponding figures were 3% and 47%, while 41% were not open to AI doing that task. These responses describe public preferences, not a clinical accuracy test.
Oversight is part of the proposed product
Pew also finds 60% uncomfortable with AI use for tasks such as diagnoses and treatments, compared with 30% comfortable. That broad comfort question and the task-specific questions use different framing. Their percentages should not be combined into a single approval score.
Franklin’s analysis is that a healthcare AI interface needs to make responsibility understandable. Showing a result with no explanation of who checks it creates a different experience from showing the same result as a draft awaiting a clinician. A human-review label is only informative when it corresponds to a real review process.
What readers should take from the evidence
The survey helps explain expectations around deployment. It does not establish that a particular system is safe, effective or suitable for an individual patient. Those questions require evidence about the system and its intended use.
For people assessing a healthcare service, useful questions concern the actual workflow: what the AI handles, who reviews its output and how an error can be corrected. The study’s strongest practical contribution is to keep those choices visible instead of treating all healthcare AI as one undifferentiated feature.