Like Father, Like Son: Kinship-Aware Preference Mapping (KARMA) for Automatic Alignment in Large Language Models