Reshaping the Happy Face Advantage with Reinforcement Learning

We systematically investigate whether reinforcement learning can modify social category biases in emotion recognition. In Experiment 1 (N = 40), we replicated that the Happy Face Advantage is influenced by social category membership. People were faster at recognizing happiness as happiness than anger as anger for White–Dutch faces, while no difference was found for Moroccan–Dutch faces. In Experiments 2–3 (Ntotal = 144), we used a reinforcement learning go/no-go task, in which people learned to act to images of Moroccan–Dutch faces to obtain rewards and to not act to images of White–Dutch faces to avoid punishments before participating in the emotion recognition task. Results show that reinforcement learning influences emotion recognition. Instead of the commonly observed interaction effect between social category and expression valence, we consistently found a main effect of valence on emotion recognition. These findings suggest that aligning (in)actions with rewards/punishment changes emotion recognition.