Evolution would condition different animals to have different homeostatic set-points (and thus tolerances for uncertainty, for example). My point about the cats was not that they exhibit both behaviors, but rather that they are an example of an animal which has generally less tolerance for uncertainty actually seeking out prediction errors in a controlled context ("curiosity") in order to minimize long-term uncertainty going forward (due to "fear"). If you are interested, this has been even modeled outside of the FEP research community in the context of Deep RL as well: https://arxiv.org/abs/1912.05510