On the underdetermination of theory by evidence