Generic generalizations play an important role in communication and learning. Recently, we argued that all generics favor stability across background circumstances. This, however, appears to be false for minority-characteristic generics (“ducks lay eggs” - only females do). In response to this challenge, we developed a hypothesis that highly interactive systems (like sexually-reproducing species) resist partitioning into subcomponents (male vs. female) in the course of generic evaluation. Consequently, a higher prevalence of the property (“laying eggs”) in one subset does not qualify as instability, and generics attributing that property to the kind are not penalized. We evaluated this hypothesis in an experiment with 99 adults, manipulating descriptions of the manner in which novel species reproduced: interactive vs. non-interactive. As predicted, generic generalizations describing interactive mechanisms were endorsed more than non-interactive equivalents. These findings support our stability argument and highlight the key role of causal-theoretical considerations and conceptual representation in assessments of generics.