We explore how adults learn counterintuitive causal relationships, and whether they discover hypotheses by revising their beliefs incrementally. We examined how adults learned a novel and unusual causal rule when presented with data that initially appeared to conform to a simpler, more salient rule. Adults watched a video of several blocks placed sequentially on a “blicket detector”, and were then asked to determine the underlying causal structure. In the “near” condition the true rule was complex, but could be found by making incremental improvements to the simple and salient initial hypothesis. The “distant” condition was governed by a simpler rule, but to adopt that rule participants had to set aside their initial beliefs, rather than revising them incrementally. Adults performed better in the near condition, despite this rule being more complex, providing some of the first evidence for an “explore-exploit” trade-off in inference, analogous to the trade-off in active learning.