A new perspective paper argues for "measuring data" as a critical task to advance responsible AI development. Just as physical objects can be measured, data used to train AI systems should also be quantitatively analyzed to understand its composition.