Generative AI And Self-Disclosure In Underprivileged Children: Evidence From Hiwel Learning Stations

Abstract

Conversational AI platforms such as ChatGPT and Gemini have become part of everyday life, with human-like interactions encouraging users to disclose personal thoughts and feelings with reduced fear of judgment. This study examined whether access to such platforms enhances self-disclosure among children and early adolescents (aged 6–14) from underprivileged backgrounds, using HiWEL Learning Stations, public, internet-enabled kiosks for self-directed learning.

A quasi-experimental design was conducted across two underprivileged communities in New Delhi, India. The experimental site provided access to ChatGPT and Gemini, while the control site retained the standard HiWEL interface without generative AI. Eighty children (40 per group) completed a self-disclosure scale at baseline and after three months. The experimental group scored significantly higher than the control group (mean difference = 63.80; t (78) = 13.873, p < .001), with self-disclosure rising 101.4% across all five dimensions: Relationships, Personal Matters, Beliefs, Interests, and Intimate Feelings. A subsequent human–computer trust assessment revealed moderate-to-high trust (M = 79.41), with competency rated highest.

These findings show generative AI can significantly enhance self-disclosure among underprivileged children in minimally supervised community learning environments, highlighting its potential for inclusive learning while underscoring the need for robust child safeguarding and AI governance.