Explain the fundamental challenge of value loading in artificial intelligence systems. Discuss how we might specify human values to AI systems without directly programming them, considering the complexity and context-dependence of human values. Analyze at least three theoretical approaches to AI alignment and evaluate their strengths and weaknesses. Finally, propose a research framework for making progress on this problem.