The problem is not the AI code, but nobody knows anything anymore

The relationship between complexity and understanding has always been central to how societies function. When a technology becomes sufficiently intricate, the number of people who can actually grasp its inner workings shrinks dramatically. This wasn't invented by AI; it's a pattern visible across nuclear physics, aviation, modern medicine, and financial derivatives. What's different now is the scale. AI systems have become so deeply embedded in everyday decisions, and the underlying mathematics so abstract, that the gap between users and creators has widened beyond what we typically see even in highly specialized fields. The practical problem emerges when no one in a room can fully explain what the system is doing or why it made a particular choice. That's a departure from most technologies where at least some subset of trained professionals understands the mechanism. The core issue touches on what researchers call interpretability or explainability. Large machine learning systems, particularly those based on deep neural networks, don't work the way traditional software does. You can't simply read the code and understand the logic step by step. Instead, these systems learn patterns from vast amounts of data by adjusting millions or billions of numerical parameters. The result works. It generates plausible text, identifies objects in images, or predicts useful patterns. But explaining exactly how any specific decision flows from the underlying parameters remains genuinely hard, even for the researchers who built the system. This isn't a failure of engineering; it's a fundamental characteristic of how these systems learn. The internal representations they develop are often alien to human reasoning. Knowledge fragmentation compounds the problem. In past eras of major technological transition, there was at least a shared baseline of understanding. Engineers, regulators, journalists, and educated citizens had overlapping mental models of how electricity worked, or aviation, or telecommunications. That common ground created a foundation for productive debate and governance. With AI, the baseline has fractured. Machine learning practitioners speak a technical dialect involving loss functions and gradient descent. Product managers think in terms of metrics and performance. Policy makers frame questions around regulation and harm. Ethicists raise concerns about bias and fairness. The public encounters outcomes without understanding the mechanism. Each group is partly right, but they're not actually speaking the same language about the same phenomenon. The shared conceptual ground has eroded. This erosion matters because it affects how societies make decisions about powerful systems. Regulation, oversight, and public trust all depend on some meaningful accountability. When nobody in the room including the people who built it can point to a clear causal chain explaining why a system made a decision about hiring, lending, or bail recommendations, the typi