Cross-Cultural Variation in Dishonesty Toward Humans and Artificial Agents Depends on Agent Pupil Size

preprint OA: closed
View at publisher

Abstract

As artificial agents become more involved in social and economic interactions that rely on honest behavior from humans, they are increasingly designed with human-like social cues, such as pupil size, to appear more approachable. Pupil size is subtle, but also a powerful cue that can guide our decisions. Yet, we still know little about how pupil size shapes people’s behavior, especially across cultures where agents and their signals are interpreted and accepted differently. In this study, 275 participants from Europe and Japan played a coin toss game where lying increased financial gain. They interacted with a human confederate, a virtual avatar, or a robot. Agents' pupil sizes were manipulated to appear large or small; for human confederates, this was achieved using custom-designed contact lenses. Our results suggest that cultures differ in how dishonest behavior is directed toward human and artificial agents: European participants were more dishonest toward artificial agents than toward humans, while Japanese participants showed no such differentiation across agent types. Beyond this, pupil size influenced dishonesty specifically towards avatars in culture-specific ways, with European being more honest towards avatar with small pupils compared to large, and the opposite pattern for Japanese participants. Exploratory analyses further revealed cultural differences in mind perception attributions across agent types, alongside overlapping pupillary responses towards agents across cultures. These findings highlight the need to consider cultural context when designing artificial agents that use human-like cues to guide social behavior: what subtly communicates warmth in one culture may signal something else in another.

My notes (saved in your browser only)

Citation neighborhood (no data yet)

We don't have any in-corpus citations linked to this paper yet. This is a recent paper (2025) — citers typically take a year or two to land, and the OpenAlex reference graph may still be filling in.

Source provenance

europepmc
last seen: 2026-05-20T01:45:00.602351+00:00