Define a comprehensive 'Constitution' for an Artificial General Intelligence system. The constitution must address the 'Alignment Problem' by explicitly weighting human rights, utilitarian outcomes, and the preservation of human agency. Create a decision-tree framework that the AI can use to resolve conflicts between these values in edge-case scenarios, such as the Trolley Problem or triage in medical resource scarcity. Explain how you would use RLHF (Reinforcement Learning from Human Feedback) to train the model to adhere to this constitution without overfitting to specific examples.