<![CDATA[Purpose – This study investigates the "AI Paradox" and the behavioral divergence between AI Augmentation and AI Replacement among 112 Grade 9 SMTP students. It addresses the critical problem of cognitive offloading, where high-quality AI-assisted assignments mask a decline in individual linguistic proficiency. The research explores how the BOT-AI Model can be used to distinguish between productive scaffolding and counter-productive dependency in English Language Teaching (ELT).Design – A 12-week quasi-experimental investigation was conducted at Petcharik Demonstration School. Participants were categorized into two groups based on a Digital Interaction Log (DIL) Matrix and the AI Literacy Engagement Scale (AI-LES). The methodology compared performance across 10 weeks of AI-assisted assignments against a strictly proctored, unassisted post-test to identify the "Paradox Gap."Findings – Results confirm a significant AI Paradox. The Replacement group (n=50) exhibited a sharp 10.70-point decline in proctored environments, demonstrating a "mirage of competence" through cognitive offloading. Conversely, the Augmentation group (n=62) maintained consistent proficiency with a large effect size (d = 1.62), utilizing AI as a constructivist scaffold with an average of 6.4 iterations per task compared to only 1.2 iterations in the Replacement group. Research implications/limitations – The findings necessitate assessment reform and a shift toward process-oriented AI pedagogy. Limitations include the specific high-achieving SMTP context within a Thai municipal school, which may vary in broader educational settings.Originality – This research provides novel empirical evidence of the AI Paradox using real-time log data. It offers a scalable framework for educators to monitor digital interaction patterns, ensuring that technology serves to amplify human intellect rather than replace it.]]>
