AI-mediated research agency formation in higher education: Autonomy, self-efficacy and innovation in early-career scientific training

Checked 30 Sept 2026. Non-commercial or no-derivatives license: full text kept internally; only metadata and abstract are indexed and used.

Generative AI is becoming part of the infrastructure of higher education research training, yet little is known about how AI-mediated environments shape early-career researchers’ agency. This study examines AI dependence, defined as reliance on intelligent tools as a primary source of cognitive support in judgement-intensive research tasks, and interprets its educational risk as epistemic delegation: the transfer of problem framing, methodological choice and interpretive authority from the researcher to the intelligent system. Drawing on an AI-mediated research agency formation framework, we examined whether research autonomy and research self-efficacy mediate the association between AI dependence and innovative research behaviour, and whether supervisory support moderates these pathways. An anonymous questionnaire was circulated through peer-based WeChat groups among astronomy doctoral students and postdoctoral researchers in China. After pre-specified data-quality screening, 420 valid responses were retained. Split-sample factor analyses supported a broad one-factor AI-dependence scale, and item-level confirmatory factor analysis supported the five-factor measurement model. AI dependence was negatively associated with research autonomy and research self-efficacy, both of which were positively associated with innovative research behaviour. The indirect association through research autonomy was stronger than that through research self-efficacy. Supervisory support was positively associated with autonomy, self-efficacy and innovation, and weakened the negative associations between AI dependence and research autonomy and innovative behaviour; the interaction predicting self-efficacy was directionally consistent but did not reach statistical significance. The findings shift the debate from AI-enabled productivity to AI-mediated researcher formation and position supervisory support as a relational foundation for epistemic supervision through explanation, verification and reflective AI use.

• Han & Liu 2026 provide survey evidence that supervisory support "weakened the negative associations between AI dependence and research autonomy and innovative behaviour," a micro-level, empirical analogue of Park et al. 2026's model result that mandatory practice preserves capability—together suggesting structured human oversight is the practical lever against AI-induced deskilling. Trace →
• Gao & Zhang 2026 specify what the "epistemic supervision" called for by Han & Liu 2026 looks like in practice: supervisors had students "mark AI-influenced parts and explain why suggestions were accepted or rejected," which "pushed [them] to connect AI outputs with source evidence, methodological assumptions, and argumentative responsibility." Trace →

• Supervision and doctoral education