Learning under compound risk vs. learning under ambiguity – an experiment