Fig. 3

People in Japan and the United States expect AI agents to be as cooperative as humans, but only in Japan they are as likely to reciprocate cooperation with both types of co-player. Top chart: proportions of participants in Japan (dark columns) and the United States (light columns) who predicted that their human (H) or AI co-player would cooperate. Bottom chart: rates of cooperation among participants who predicted that their co-player would cooperate. Not all participants who were assigned to the role of player two and made a prediction about their co-player’s choice in the Trust game had an opportunity to make a choice themselves. (This was conditional on the first player’s decision in the game.) Bars: mean ± 1 s.d.. *, ***: p < 0.05, p < 0.001 in Pearson’s chi square one-sided tests for difference in proportions. Below chart: the number of cooperative choices (n) and the total number of observed choices (N) in each treatment.