Research on active category learning—i.e., where the learnermanipulates continuous feature dimensions of novel objects andreceives labels for their self-generated exemplars—has routinelyshown that people prefer to sample from regions of the space withhigh class uncertainty (near category boundaries). Prevailingaccounts suggest that this strategy facilitates an understanding of thesubtle distinctions between categories. However, prior work hasfocused on situations where category boundaries are rigid. Inactuality, the boundaries between natural categories are often fuzzyor unclear. Here, we ask: do learners pursue uncertainty samplingwhen labels at the boundary are themselves uncertain? To answerthis question, we introduce a fuzzy buffer around a target categorywhere conflicting labels are returned from two ‘teachers,’ then weevaluate how sampling and representation are affected. We find that,relative to the rigid boundary case, learners avoid uncertainty,opting to sample densely from highly certain regions of the targetcategory as opposed to its boundary. Subsequent generalization testsreveal that the sampling strategies encouraged by the fuzzyboundary negatively affected participants' grasp of categorystructure, even outside the fuzzy buffer zone.
