We introduce a modification to the two-timescale games studied in the evolution of preferences (EOP) literature. In this modification, the strategic process occurring on the long timescale is learning by an individual across his or her lifetime, not natural selection operating on genomes over multiple generations. This change to the longer timescale removes many of the formal difficulties of EOP models and allows us to show how two-timescale games can provide endogenous explanations for why humans sometimes adopt interdependent preferences and sometimes exhibit logit quantal response functions. In particular, we show that our modification to EOP explains experimental data in the Traveler's Dilemma. We also use our modification to show how cooperation can arise in nonrepeated versions of the Prisoner's Dilemma (PD). We then show that our modification to EOP predicts a "crowding out" phenomenon in the PD, in which introducing incentives to cooperate causes players to stop cooperating instead. We also use our modification to predict a tradeoff between the robustness and the benefit of cooperation in the PD.