Infants' reasoning about samples generated by intentional versus non-intentional agents.

The current experiments investigate how infants use goal-directed action to reason about intentionally sampled outcomes in a probabilistic inference paradigm. Older infants and young children are flexible in their expectations of sampling: They expect random samples to reflect population statistics and non-random samples to reflect an agent's preferences or goals (Kushnir, Xu, & Wellman, 2010; Xu & Denison, 2009). However, more recent work shows that probabilistic inference comes online at approximately 6 months (Denison, Reed, & Xu, 2013; Kayhan, Gredebäck, & Lindskog, 2017; Ma & Xu, 2011; Wellman, Kushnir, Xu, & Brink, 2016), and thus, these sampling assumptions can be investigated at the age probabilistic reasoning first emerges. Results indicate that 6-month-old infants expect a human agent to sample in accord with their goal and do not expect the same of an unintentional agent-a mechanical claw. By 9.5 months, infants expect the mechanical claw to sample in accord with random sampling. These results suggest that infants use goals to make inferences about intentional sampling, under appropriate conditions at 6 months, and they have expectations of the kinds of samples a mechanical device should obtain by 9.5 months.