Evidence against empiricist accounts of the origins of numerical knowledge