INTERCULTURAL DIFFERENCES IN YOUTH COMMUNICATION WITH AI

Abstract

Artificial intelligence (AI) has become part of everyday life, and young people increasingly use AI tools for study, work, and daily problem solving. At the same time, research highlights the risk of harmful or excessive reliance, including potential reductions in critical thinking and the emergence of dependency-like patterns. Understanding how youth conceptually position AI in relation to human social partners is therefore important for trust calibration and for anticipating how the human–AI relationship may reshape social interaction.The aim of this study was to map how youth in two cultural contexts (Ukraine and Italy) perceive AI within a semantic space and to compare AI positioning with two socially salient human targets (a friend and an enemy). Two groups of participants (Ukraine: n = 63; Italy: n = 63; ages 18–30, M = 22) completed online ratings of AI, a friend, and an enemy on a semantic differential with 24 bipolar adjective pairs (1 = left pole; 7 = right pole). Principal component analysis (PCA) with varimax rotation was performed separately for each target in each cultural group (six solutions). Components were retained using Kaiser’s criterion ( λ >1), and salient loadings were defined as | f | ≥ 0,40. Across cultures, AI evaluations were organized by separable socio-moral and instrumental dimensions, but the dominant meaning of axes differed. In the Ukrainian sample, the leading AI component emphasized trust and safety, whereas in the Italian sample the dominant structure reflected a calmer and more manageable representation. In both groups, AI remained distinct from human targets due to dimensions capturing artificiality and reduced emotionality. Friend and enemy structures showed coherent contrasts between support/resourcefulness and moral threat, with culture-specific integration of competence-related devaluation in the enemy's representation. The study is limited by modest sample sizes and its cross-sectional design future work should test the stability of these semantic structures in larger and more diverse samples, incorporate contextual factors (e.g., quality of offline social life, social status, strength of friendship networks), and examine whether individual differences (e.g., emotional intelligence and the strength of social ties) buffer risks associated with intensive AI use.