We develop a model of spatial elections that departs from the standard model in three important respects. Our parties' information of voters' preferences is limited to polls; our parties can be either office-seeking of ideological; and our parties are not perfect optimizers, i.e. they are modeled as boundedly rational actors. Since our imperfect parties do not necessarily find optimal positions, rather than concern ourselves with existence and location of equilibria, we trace the trajectory of winning party positions. The outcomes are subsequently evaluated with respect to a measure of social welfare, centrality. Our results suggest that in fair voting systems, two party elections lead to normatively appealing outcomes. We are seeking to introduce the role of computers genearlly and adaptive artificial agents (AAA) specifically, to the study of parties, voters, and elections in spatial models. We argue for using adaptive parties to add behavioral complexity to stand formal models of politics without sacrificing a logical foundation. Doing so may not only revise our judgments about the relevance of spatial voting models to real elections, but it may lead us to important insights about what occurs in those elections.