Toward Cultural Competence in Social Robotics

In human behavioral studies, cultural competence refers to the ability to recognize, understand, and adapt to diverse cultural contexts. This concept can be extended to artificial systems such as social robots (SRs). In this domain, the absence of cultural competence may lead to inappropriate behaviors, reduced user satisfaction, system failures, and ethical concerns. Although the field has received increasing scholarly attention, several critical challenges remain insufficiently addressed. This thesis investigates the technical and ethical implications of the underrepresentation of minority cultures in machine learning (ML) models deployed in SRs. Specifically, it examines how culturally imbalanced datasets contribute to the development of culturally incompetent systems that fail to generalize to, or adapt effectively for, users from underrepresented cultural groups. To mitigate these limitations, this work evaluates bias-reduction strategies based on multi-task learning and data preprocessing techniques. Furthermore, acknowledging that cultural competence is often conceptualized primarily through the lens of national culture—a practical but inherently reductive approximation—this thesis proposes and evaluates a novel architecture for cultural adaptation. The proposed framework accounts for multiple cultural dimensions beyond nationality, addressing the limitations of relying exclusively on conventional ML models to achieve dynamic and individualized cultural adaptation.