On this page

Synthesis: A survey of 381 full-time students at a technical university in Uzbekistan compared how undergraduates and graduate students report using generative AI alongside attention, cognitive load, motivation, self-regulation and ethical autonomy. Graduate students scored higher on six of seven scales, most sharply on ethics and academic autonomy (d = 1.47) and AI use (d = 1.29); cognitive load did not (p = .296). Ethics (β = 0.446), self-regulation (β = 0.279) and motivation (β = 0.178) predicted the authors' exploratory Neuro-AI Pedagogical Competence scale (R2 = 0.799), but use frequency did not (β = 0.030, p = .399). Academic stage appears to shape how students regulate and justify AI use, not just how often they use it, though overlapping scales keep the model exploratory.

Key Findings

  1. Graduate students reported far heavier and far more uniform AI use than undergraduates. The narrow spread suggests habitual, purposeful use rather than occasional experimentation.
  2. Ethical awareness and academic autonomy produced the widest gap between the two groups. Undergraduate scores were uneven, not uniformly low.
  3. Self-regulation and Neuro-AI pedagogical competence moved together, with graduate students ahead on both.
  4. Perceived cognitive load was the one construct that did not differ, so felt overload looks like individual variation rather than a property of program stage.
  5. Attention was the weakest area across the whole sample, and graduate students reported fewer attentional difficulties than undergraduates on the reverse-coded scale.
  6. Use frequency did not survive as a predictor once ethical awareness entered the model, which the authors treat as evidence that judgment matters more than exposure.

A level gap in judgment, not in felt workload

The pattern is distributional, not just about means. Graduate standard deviations on ethics, self-regulation and competence were roughly a third of the undergraduate values, with interquartile ranges compressed into the top of the scale while undergraduates reached the minimum. That heterogeneity matters more than the average: a bachelor's cohort holds students who appraise AI output critically alongside those who report almost none of that disposition.

The cognitive load result is easy to misread as a failure. The scale did not track academic level (p = .296), which the authors read as overload following individual variation rather than program stage; they also concede that it blends perceived overload with AI-assisted understanding, limiting how far the null can be pushed.

Why frequency of use drops out of the model

Students who use AI tools more often report stronger metacognitive and ethical profiles, yet once those are in the model, frequency adds nothing. The authors read this through self-determination theory: internalized regulation and ethical commitment, not exposure, co-vary with competence.

The reading needs caution: predictors and outcome come from one self-report instrument with overlapping items. Two items, one on the AI Use scale and one on the competence scale, are near-identical about organizing academic work with AI, and the authors flag this overlap as a likely contributor to R2 = 0.799. What the study supports is descriptive: self-reported AI literacy in the broad sense clustered with self-regulation and ethics, not usage intensity. It does not support a causal claim that teaching ethics raises competence.

An instrument still under construction

The competence scale is both contribution and weak link. The prefix signals awareness of one's own attention, memory and load during AI work, not neurological measurement; no neurophysiological data were collected. Of the six items retained, only one directly references attention, memory and learning characteristics; the rest restate AI use and study regulation captured elsewhere. Discriminant validity is unestablished and no confirmatory factor analysis was run.

The reliability reporting is transparent: the 30-item instrument spans seven scales with alpha from 0.717 to 0.895, four items were dropped after principal axis factoring, and retained loadings run from 0.622 to 0.930. As a first operationalization of academic responsibility it is useful; as a validated construct it is not there yet.

What this means for practice

  • Instructors: Treat ethical reasoning as a teachable undergraduate target: undergraduates averaged 3.527 (SD = 1.003) against 4.627; build appraisal tasks around the line between support and substitution.
  • Instructors: Do not read heavy AI use as competence. Frequency predicted nothing once self-regulation and ethics were controlled, so require planning, monitoring and justification routines, not prompt counts.
  • Designers: Differentiate AI-integrated tasks by level: scaffolded, structured tasks early, more autonomous ethics-focused tasks at postgraduate level.
  • Administrators: Fund attention and workload support separately from tool access: digital attention was the lowest scale (M = 2.642) and cognitive load did not differ by level, so buying more licenses fixes neither.
  • Researchers: Treat the scale as a draft: run confirmatory factor analysis on an independent sample and drop items duplicating the predictors before comparing cohorts.

Limitations

  • Cross-sectional, self-report, single institution: 381 full-time students at one technical university over one semester, compared on a balanced subsample of 302.
  • Construct overlap inflates the regression: R2 = 0.799 is association among overlapping scales, not a differentiated model.
  • Scale weaknesses: cognitive load mixes overload with AI-assisted understanding, ethics has three items with no behavioral validation, and one attention item needed reverse coding.
  • Social desirability likely inflates self-reported ethics on a topic as sensitive as academic integrity, and no performance measure anchors any scale.
  • The methods section reports exemption from formal ethics board approval while the ethics statement reports approval; either way, independent instrument review is undocumented.

Citation

Zaripova, D., Zohirov, K., Rasulova, N., Khayitov, U., Saidova, D., Temirov, Z., & Nasimov, R. (2026). Generative AI use, self-regulation, and Neuro-AI Pedagogical competence in higher education: a comparative study of undergraduate and graduate students in Uzbekistan. Frontiers in Education, 11, 1846709.

Embed this page

Copy the code below to embed this page in a learning management system or other website. The embedded version shows only the page content — it hides the site header, navigation, and footer.