You don't need to solve every problem to make AI worth the investment, just a large enough handful of billion-dollar problems.