Would you Rather? A New Benchmark for Learning Machine Alignment with Cultural Values and Social Preferences