Language modelling is the objective (if we ignore RLHF). That doesn't mean that interesting kinds of reasoning can't emerge. You could just as easily dismiss humans as being 'biological replicators' which just 'reproduce their genomes'.