The DI methodology for building human-governed AI systems: train humans first, codify their knowledge into the AI, feed AI outputs back into human understanding, then build the AI to augment humans while humans govern it.