BEYOND ACCESS:

Abstract

Through a survey of 513 university students in a criminal justice and security LSP course, this study investigates the effects of LLMs on language learning and instruction. Four hypotheses on motivation, interaction quality, critical assessment, and integration are tested using non-parametric techniques. Compared to peers who perceive AI as neutral or harmful, students who believe it eases learning report far greater enthusiasm and active involvement (p < 0.001). Interaction quality, prompt effectiveness, and comprehension of AI output are considerably greater among students using AI proactively and autonomously rather than responding to teacher-directed prompts (p < 0.001). No discernible difference emerged in self-reported grammatical correctness between those who critically check AI replies and those who accept them without question (p = 0.72), indicating verification alone does not ensure increased accuracy. Deeper AI integration shows a favorable correlation with overall satisfaction (rho = 0.529) and perceived efficacy (rho = 0.543), both significant at p < 0.001. These results suggest learner satisfaction and engagement in AI-supported language learning are driven by learner agency and perceived utility rather than mere access to AI technologies. Proactive, self-directed AI use may improve interaction quality and motivation more effectively than teacher-controlled, prescriptive adoption. The findings add empirical support to discussions on the pedagogical significance of LLMs while cautioning that critical assessment skills may need pairing with explicit accuracy-focused teaching to produce measurable grammatical gains. Implications for task design, evaluation, and teacher preparation in LLM-integrated courses are examined, offering curriculum designers recommendations for balancing autonomy with pedagogical scaffolding.