We investigate the implementation of social choice functions (SCFs) from an epistemological perspective. We consider the possibility that in higher-order beliefs there exists an honest agent who is motivated by intrinsic preference for honesty as well as material interest. We assume weak honesty, in that an honest agent is mostly motivated by material interests and even tells white lies. Importantly, this study assumes that “all agents are selfish” never happens to be common knowledge. We then show the following positive results for the implementability: In complete information environments, with three or more agents, any SCF is uniquely implementable in the Bayesian Nash equilibrium (BNE). In asymmetric information environments, with a minor restriction named information diversity, any incentive-compatible SCF is fully implementable in BNE. An SCF, whether material or nonmaterial (ethical), can be implemented even if all agents are selfish and “all agents are selfish” is mutual knowledge.