This paper examines the psychological differences between hedonic and utilitarian patterns of preference behavior. Instead of using latent variables like self-control and emotion to explain these differences, we show that they emerge as natural consequences of solving two different but related problems within an inductive framework of preference learning. We show that hedonic decisions involve tracking the variability of a binary variable, whereas utilitarian decisions require the maintenance of a distribution over a vector of object labels. Computational experiments show that this difference in cognitive representation ensures that hedonic decisions have a lower cognitive sampling cost, which makes them less effortful. Further experiments reveal differences in error rates as a function of deliberative effort between the two paradigms. Deliberative effort benefits utilitarian choices, but not hedonic ones. Overall, our work demonstrates the critical role of cognitive representations in extracting strikingly different behavior patterns from simple models of information processing.