Language, logic, and the future of AI: An early-Wittgensteinian perspective — LessWrong