A questionnaire survey on context-dependent requirements for the trustworthiness of artificial intelligence and humans

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Trustworthiness has become an important focus of artificial intelligence (AI) research. Although trustworthiness is a multifaceted concept and its requirements may depend on the context in which AI is used, little is known about how the public perceives these requirements. To address this issue, we conducted a questionnaire survey with nonexpert participants to examine the requirements that the public considers important for trustworthy AI and how their perceived value differs between social and nonsocial contexts. We further compared these context-dependent patterns with those observed for human trustworthiness. Our findings showed that, across social and nonsocial contexts, confidentiality and privacy protection were among the most important requirements for AI trustworthiness whereas the relative importance of other requirements depended on the context. In nonsocial contexts, transparency and interpretability of decision-making were particularly important whereas in social contexts, the ability to collaborate and communicate and a positive public image and attractive appearance were predominantly significant. Although similar context-dependent patterns were observed for human trustworthiness, findings consistently highlighted the ability to collaborate and communicate as a requirement for trusting humans, regardless of context, in contrast to AI trustworthiness. These findings provide insights into the public’s perception of trustworthy AI and highlight the importance of considering the context in which AI is used when examining AI trustworthiness requirements.