Data Representativeness Through a Sociocultural Lens in AI Modeling for Mental Health

Abstract

Swati S. Karve and Prachee P. Joeg

Background: Artificial Intelligence (AI) is reshaping mental health diagnosis and treatment. However, most AI models are built on Western-centric datasets, limiting their applicability in culturally diverse settings like India. Socio-cultural realities need to inform datasets, and this review seeks to understand the challenges in making data more culturally nuanced and representative. This paper aims to identify the socio-cultural aspects that affect data representativeness in AI modeling in anxiety and depression in the Indian context. A literature survey was conducted of studies from 2000– 2024 focusing on AI in mental health and the socio-cultural dimensions of anxiety and depression in India based on the exclusion and inclusion criteria that were defined for the survey. PRISMA guidelines are used to report these criteria.

35 studies out of 101 studies reviewed were included in the final review. Two major themes emerged: (1) The socio- cultural context in the manifestation of anxiety and depression in India, specific anxiety and depression symptoms in India and idioms of distress with respect to anxiety and depression in India and (2) Missed opportunities in data collection due to stigma, lack of screenings in public health, and underrepresentation certain groups and absence of evidence-based treatments. The implications of these themes in terms of data representation in AI modeling are discussed. The review informs practitioners, AI developers, and end-users about the critical issue of representativeness bias within socio-cultural frameworks in mental health.

PDF