Inference of Causal Effects when Adjustment Sets are Unknown