Should AI systems prioritize explainability or raw accuracy?

explainability without understanding is useless tbh. accuracy is what actually matters unless being wrong kills people