Empathy for the Devil: Using Counter-Stereotypical Deepfakes to Manipulate Affective Polarization and Democratic Attitudes

Affective polarization has escalated in numerous countries, raising concerns about its detrimental effects on democratic norms. In response, scholars have developed experimental interventions that, while successful in reducing polarization, have proven mostly ineffective at influencing democratic attitudes. Focusing on Brazil’s polarized landscape, this study pioneers the benign use of deepfake images—fabricated images synthesized by deep learning algorithms—in political research, promoting prosocial and democratic attitudes. I introduce the use of counter-stereotypes into the research on polarization, documenting that out-group representations that contradict prevailing stereotypes foster empathy substantially. Furthermore, this reduction in animosity has downstream effects on support for initiatives safeguarding free and fair elections, substantiating the promise of diminishing polarization to reinvigorate democracies. Images likely amplify treatment effects, and this article also discusses whether their inclusion violates the information equivalence assumption, an assessment with important implications for the design of experimental research.