Direct-Answer Versus Socratic Generative AI Support in Programming Learning: Differences in Epistemic Laziness, Computational Thinking, and Computational Thinking Network Structures

Generative artificial intelligence can support programming learning, but differences associated with response modes remain unclear. This two-class quasi-experiment compared direct-answer support (n = 57) with Socratic support (n = 48) across three 90-minute sessions in a Java course. Pre- and post-intervention questionnaires assessed epistemic laziness and five computational thinking dimensions, while reflection logs were analyzed using epistemic network analysis. After adjustment for corresponding pretest scores, Socratic support showed lower self-reported epistemic laziness and higher abstraction, decomposition, algorithmic thinking, and evaluation scores, with no significant difference in generalization. Network analysis revealed distinct computational thinking configurations: direct-answer support showed stronger decomposition–algorithmic-thinking and algorithmic-thinking–evaluation connections, whereas Socratic support showed stronger abstraction–decomposition, abstraction–generalization, and decomposition–generalization connections, suggesting different patterns of problem analysis, solution development, and knowledge transfer. Given the two-class design, findings are interpreted as between-case differences. Overall, response mode may shape computational thinking outcomes and the organization of related processes in programming learning.