Pew survey finds Americans want human review of healthcare AI

Pew Research Center released a new report on October 29 showing that Americans' openness to healthcare AI depends on the task and whether a person reviews the system's work. The report draws on interviews with 3,488 U.S. adults conducted June 22–28. October 29 is the release date of these findings, not the date the interviews took place.

Pew found that 60% were uncomfortable with AI being used in their own healthcare for tasks such as diagnosis and treatment, compared with 30% who were comfortable. That broader question should be kept separate from task-specific questions about an AI system working alone or under human review. A general discomfort measure does not mean that respondents rejected every possible use.

FingerLakes1's separate coverage emphasizes that the research measures attitudes rather than medical performance. It describes stronger acceptance of translation and human-reviewed assistance than of autonomous decisions. The report therefore adds evidence about what people want from healthcare systems, without demonstrating whether a particular AI tool is accurate or safe.

The methodology matters when interpreting the findings. Pew used its probability-based American Trends Panel, with SSRS conducting interviews online and by telephone in English and Spanish. Results were weighted to account for selection and nonresponse and to align with population characteristics. The reported full-sample margin of sampling error is plus or minus 1.8 percentage points; subgroup results have different precision.

The methodology also notes that question wording and the practical difficulties of surveys can introduce error or bias beyond sampling uncertainty. These are U.S. adult opinions collected in June, rather than a live measure of attitudes after every subsequent model release or healthcare announcement.

For clinicians and health systems, the central implication is to define the proposed role of AI clearly. Administrative assistance, support for a clinician and an autonomous decision are different arrangements. Public acceptance can inform how those arrangements are discussed, but it cannot substitute for clinical evaluation, privacy safeguards or evidence about patient outcomes. Those questions require their own research.