In this work, we address the problem of how altruistic behavior evolves in social systems by studying the evolutionary dynamics of the ultimatum game in complex networks. In the ultimatum game two players decide how to distribute a reward: one makes an offer and the other responds by accepting or rejecting it (both players receiving nothing in the latter case). Then, fair (or altruistic) behavior corresponds to offers close to a half of the total reward. Specifically, we consider two types of players: (a) empathetic, whose aspiration level and offer are equal and (b) pragmatic, who do not distinguish between the different roles and aim to obtain the same benefit in any case. Evolution is driven by imitation of the strategies of neighbors with better results. We analyze numerically the asymptotic behavior of pure populations on different topologies as well as the outcome of the game when both types of players are present. The results show that pragmatic players in most situations take over the population of empathetic individuals and that social diversity, as given by heterogeneous networks, introduces new dynamical features. In particular, we show that the most connected individuals are able to impose its initial strategy to the rest of the players, with an increasing probability the closer their initial offer is to fairness. Our results point out that fairness is the most natural outcome for the evolution of the strategies in the population under quite general dynamical and topological conditions.