I argue for the plausibility of the predictive processing framework over the standard bottom-up model of perception, especially in the context of efficiently processing high-dimensional multimodal inputs, where the qualitative space of each modality has unique dimensionality and structure. 