Designing behavioral interventions with explainable artificial intelligence

Abstract: Identifying how language drives human behavior is crucial, both practically to influence people and theoretically for understanding psychology, culture, and myriad behaviors predicated on verbal or written communication. Here, we present a method based on explainable artificial intelligence (AI) that identifies what aspects of the language people read are associated with their reactions. As an example, we use this method to redesign choice tasks to modulate people’s risk taking across five novel datasets containing 38,911 responses to 255 distinct decision tasks. Overall, we both i) design interventions to influence behavior at scale with greater efficiency than methods such as AB-testing and ii) extract previously unknown drivers of behavior as a basis for developing new theories across academic disciplines.

Keywords: explainable AI, risky decision making, interpretability