In human vision, representational dimensions continually expand to enable abstraction, whereas artificial networks undergo late-stage collapse that may constrain their capacity for flexible generalization.