Goal-Direction for Simulated Agents — AI Alignment Forum