Proceedings of the Annual Meeting of the Cognitive Science Society

People readily infer how much another agent cares about their welfare, for example after observing this agent give them some of what they have. However, these inferences become more difficult when there is uncertainty over the resources someone has to share, a common real-world scenario. We develop a Bayesian com- putational model of how people infer the welfare trade-off ratio (WTR) of another agent under resource uncertainty. This model predicts that under uncertainty people should average over differ- ent possible hypotheses about their partner's resources. In a be- havioral study (N = 129), we found that across donation amounts participants' WTR estimates closely tracked those made by our Bayesian model. Notably, participants exhibited the following paradoxical pattern predicted by our model: they sometimes saw agents as less generous when they gave them more money, in cases where a high donation revealed that the agent is rich and is giving a comparatively small portion of their resources. These findings demonstrate that people are able to make sophisticated inferences consistent with rational Bayesian reasoning, updating their beliefs about a partner's resources and adjusting their WTR estimates accordingly.