Lukeprog from LessWrong proposes a strategy for tackling hard problems like Friendly AI: "Hack away at the edges", making incremental progress by exploring related subproblems and drawing connections between different fields. via LessWrong.