Search Results

ABSTRACT

This study examined whether generative AI-driven environmental education changes university students’ social norms and environmental awareness differently from conventional instruction when course content is held constant. Environmental awareness was operationalized as nature relatedness. A pre-test post-test control group quasi-experimental design was used with 257 undergraduate students enrolled in a green building environment course, 129 in the experimental group and 128 in the control group. The experimental group reached the course material through a custom ChatGPT module restricted to that material, while the control group received conventional instruction. The intervention lasted 14 weeks. Social norms were measured with a purpose-developed 10-item instrument distinguishing descriptive from injunctive norms, and nature relatedness with the 21-item nature relatedness scale. Data were analyzed with Bayesian paired samples t-tests and Bayesian ANCOVA. Both conditions produced large pre-test to post-test gains on all five dimensions, with Bayes factors above 10 to the power of 42 and effect sizes between d = 0.90 and d = 1.41. After pre-test scores were controlled, the evidence favored the absence of a group difference on every dimension: Bayes factors for adding group to the pre-test model ranged from 0.21 to 0.48, and the strongest evidence against a group effect was obtained for nature experience, BF10 = 0.24, the dimension on which an advantage for continuous access was most plausible. Men scored slightly higher than women on the nature-self dimension. The findings indicate that the quality of course content, rather than the medium through which it is delivered, accounts for the change observed here.

Keywords: generative AI, ChatGPT, environmental education, social norms, nature relatedness, Bayesian analysis