The Effects and Boundaries of Human–AI Collaborative Creation in Language Aesthetic Education: Quasi-Experimental Evidence from 10 Higher Vocational Colleges in the Chengdu–Chongqing Region (Note 1)

Abstract

This study explored the facilitating effect and application scope of human-AI collaborative creation on higher vocational students’ language aesthetic literacy to provide evidence-based support for AI-empowered language aesthetic education in higher vocational institutions. A quasi-experimental design was used, and the subjects were 600 first-year students from the ten higher vocational colleges in the Chengdu–Chongqing area. A three-group (human-AI collaborative / pure-human / pure-AI) pre-test + post-test + delayed post-test experiment was conducted. The four-dimensional, 26-item self-developed *Higher Vocational Students’ Language Aesthetic Literacy Scale* (Cronbach’s α = 0.89, CFI = 0.93, RMSEA = 0.06), the AI Self-Efficacy Scale (AISE), the AI Anxiety Scale, the work evaluation scale and the human-AI collaborative creation log were used in this study to systematically examine the impact of language aesthetic education in human-AI collaborative creation and its moderating mechanisms and application scope. The results were as follows: (1) The human–AI collaborative group exhibited significantly higher post-test total scores in language aesthetic literacy than the pure-human group (Cohen’s d = 0.72) and the pure-AI group (d = 0.85), with effect sizes for the aesthetic perception and aesthetic appreciation dimensions exceeding those for aesthetic creativity (d > 0.80 vs. d = 0.68). (2) Collaborative depth exhibited an inverted-U curvilinear effect; moderate collaborative depth (AI-assisted ideation + polishing; d = 0.65 vs. low collaboration) represented the optimal range, whereas high collaborative depth (+ evaluation intervention) resulted in attenuated effects (d = 0.58 vs. moderate collaboration). (3) AISE positively moderated the relationship between collaborative depth and aesthetic education effects (β = 0.15, p = 0.020), whereas AI anxiety negatively moderated this relationship (β = –0.12, p = 0.039). (4) Technical boundaries, cognitive boundaries, ethical boundaries, and aesthetic boundaries constituted a four-dimensional boundary framework for human–AI collaborative creation in language aesthetic education, systematically demarcating the operable space and non-negotiable bottom lines for AI intervention. Human-AI collaborative creation can effectively promote higher vocational students’ language aesthetic literacy, but the extent of this effect depends on how deeply and in what way it is coordinated with the learners’ psychological characteristics. A moderate level of collaboration depth (ideation + polishing) is ideal; thus, one must be mindful of AI anxiety interventions and the joint management of the four-dimensional boundary. The “optimal collaborative depth model” and the “four-dimensional boundary framework” built in this study provide a theoretical basis and a practical path for AI-enhanced language aesthetic education in higher vocational schools, and they have good generalizability.