Carlos Sosa Colindres

Presentation Title

How To Learn Mate Preferences

Abstract

Each of us has a unique set of mate preferences; but how we come to these mate preferences is mysterious. In prior work we have found that a model that iteratively adjusts mate preferences in response to feedback can reproduce individual preferences. However, this prior model is limited because it does not have the ability to learn in real-time. Reinforcement learning algorithms, in contrast, allow online learning from the mating market. Here we test how well a reinforcement learning algorithm—specifically the e-greedy, UCB, and Thompson Sampling strategies—approximates how people learn their mate preferences. In this model, agents attend to the mate value and interest expressed by other agents and from this, they estimate the expected value of adjustments to their preferences. Even with no prior knowledge of participants’ (n = 1044) preferences, this model reproduces individual participants’ mate preferences with respectable accuracy. This research suggests reinforcement learning provides promising models for solving the critical but underexplored adaptive problems in the ontogeny of mate preferences.