Explain the theoretical underpinnings of the AI alignment problem. Discuss the concept of instrumental convergence and the orthogonality thesis. Provide a theoretical framework for solving the value-loading problem without relying on current technical limitations.