Are Gibbs-type priors the most natural generalization of the Dirichlet process?