As artificial intelligence proliferates rapidly, understanding its impact on social interaction patterns becomes critical. Using national survey data and gift money expenditure as a proxy for social interaction, the study employs instrumental variable methods to identify the causal effect of AI use on individual social behavior. The study documents three key findings. First, AI use significantly reduces traditional social interaction through substitution effects, with instrumental variable estimates showing that OLS substantially underestimates the true magnitude, confirming the technology substitution hypothesis. Second, diminished social willingness serves as a key mediating mechanism—AI use reduces social behavior by weakening non-family social preferences, demonstrating how technology shapes behavior through preference channels. Third, the substitution effect exhibits significant demographic heterogeneity, with younger, more educated, and higher-income individuals displaying greater sensitivity to technology adoption, consistent with digital divide patterns. These findings provide micro-empirical evidence of social relationship transformation in the digital era. The results suggest policymakers should emphasize social inclusiveness in AI adoption while promoting balanced development between digital innovation and traditional social engagement.