Some experts believe the alignment problem can be fixed by making AI more human — but others say that will just make things much worse.