I've seen elsewhere it was suggested that whereas humans will want to get used to the question style of each category and learn from each one as they progress to harder answers, Watson has no such issues with jumping straight for high-value questions in all categories.