If that point about datageddon is true (strikes me as logical but im no data scientist), it seems that learning algorithms that can deduce the importance of datapoints based on a large but fuzzy dataset would be of significant importance.