ScholarWorks@성균관대학교

초록

AI conversations are often perceived as anonymous and private spaces that encourage candid self-expression, yet they can also increase vulnerability to offensive and toxic language use. This may be amplified in voice-based interactions, reducing users' awareness of their own language behavior. While research has addressed offensive language detection and AI responses, little is known about whether such approaches actually influence users' language use or self-awareness. We conducted semi-structured interviews and a 4 × 2 experimental study to address this gap. Interviews identified contextual triggers of hostile language use in AI conversations and user characteristic differences. The experiment manipulated toxic utterance re-presentation during voice-based AI debates across age groups. Results show that higher toxic utterance re-presentation led to increased self-awareness and behavioral change with distinct age-related patterns. This study contributes to understanding offensive language use mechanisms in human-AI interaction and suggests design insights for enhancing self-awareness.

키워드