In this paper, we explore the consequences for optimality of a social planner adopting two different welfare criteria. The framework of analysis is an overlapping generations model with physical and human capital. We first show that, when the social welfare function is a discounted sum of individual utilities defined over consumption per unit of natural labor, the precise cardinalization of the individual utility function becomes crucial for both the characterization of the social optimum and the policies that support it. Also, decentralizing the social optimum requires an education subsidy that is definitely positive, but its size depends in a determinant way on the aforementioned cardinalization. In contrast, when the social welfare function is a discounted sum of individual utilities defined over consumption per unit of efficient labor, the precise cardinalization of preferences becomes irrelevant. More strikingly, along the optimal growth path, the education subsidy is negative, i.e., the planner should tax rather than subsidize investments in human capital.