What does it mean when people say "AI alignment is unsolved"? I tried to find out by writing about the biggest obstacles to perfectly aligning AI models and systems with human goals, preferences, values, rules, and intentions.