Speaker
Description
Self-report surveys may not fully capture how students actually use generative AI. This research-oriented presentation compares questionnaire scores for AI-mediated autonomy and AI dependence with linked ChatGPT interaction logs from 80 EFL learners. It identifies systematic mismatches, such as learners who report high autonomy but show AI-first task onset or outsourcing of idea generation. Implications are discussed for construct validity in AI research and for classroom guidance that supports critical, responsible AI use.
Abstract section 3: Content/method
A questionnaire was administered to 135 first- and second-year engineering students. Exploratory factor analysis supported a two-factor structure consisting of AI-mediated autonomy and competence and AI dependence. For a subset of 80 learners, ChatGPT interaction logs used for preparing an English presentation task were linked to survey responses. Logs were coded for task-onset behavior, prompt intent, and revision ownership. Survey-based profiles were compared with log-based patterns.
Abstract section 1: Relevance
Surveys are widely used to examine learners’ generative AI use in EFL learning, yet self-report may not accurately represent actual behavior. This limitation is particularly important when interpreting AI-mediated learner autonomy and psychological dependence as learner characteristics.
Abstract section 4: Outcomes/results
Preliminary analyses indicate noticeable gaps between self-report and behavior. For example, some learners with high autonomy scores still defaulted to AI at task onset, while others with low dependence scores frequently relied on AI for idea generation. These findings suggest that autonomy and dependence cannot be inferred from self-report alone and support the use of mixed evidence combining questionnaires with behavioral trace data. The presentation concludes with implications for research validity and for instructional support that targets critical, responsible AI use.
Abstract section 2: Contribution/research questions
(1) Do survey-based autonomy and dependence profiles correspond to AI use behaviors observed in ChatGPT logs? (2) What recurring mismatch patterns emerge between reported profiles and logged behavior during task preparation?
| Title | Survey vs ChatGPT Logs: Gaps in EFL Learners AI Use |
|---|---|
| Teaching Context | College and university education |