I think he's implying that it requires the same basic computer science hacking that is required for all of the "cat pictures" projects he points out. The point isn't that we should choose between data-driven and not data-driven; it's that we should choose between cat pictures and the human genome.