Student-agent interaction produced far more help-seeking but a non-uniform cognitive distribution dominated by lower-order engagement (54.52%), unlike the balanced distribution of student-instructor interaction¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? In traditional learning students rarely asked for help (102 records), whereas in agent interaction help-seeking was frequent (17,856 records). → Li 2026 q2 i? Cognitive engagement in the agent context was non-uniform (Lower order = 54.52%, Medium order = 41.56%, Higher order = 3.92%), versus a uniform distribution in traditional learning. → Li 2026 (2)

Evidence¶

Li 2026¶

Li, X., Liu, Z., Jiang, S., Chen, J., & Chen, W. (2026). The impact of an LLM-based educational agent on learning achievement, cognitive dynamics, and student perceptions in computer science education. International Journal of STEM Education. https://doi.org/10.1186/s40594-026-00641-y

q2 · i? · causal · r2

Descriptive statistics over coded interaction logs from both paradigms: 102 student-instructor records versus 17,856 student-agent records, showing students "exhibit frequent help-seeking behaviors" with the agent. Coding achieved Cohen's Kappa κ = 0.79.

"In traditional learning, they rarely ask for help (cognitive code = 102), whereas in human-DBagent interaction learning scenarios, students exhibit frequent help-seeking behaviors (cognitive code = 17856)."

Li 2026 (2)¶

Li, X., Liu, Z., Jiang, S., Chen, J., & Chen, W. (2026). The impact of an LLM-based educational agent on learning achievement, cognitive dynamics, and student perceptions in computer science education. International Journal of STEM Education. https://doi.org/10.1186/s40594-026-00641-y

q2 · i? · causal · r2

Distribution of Bloom-based cognitive codes across interaction paradigms shows agent-context engagement was "non-uniform", with lower-order at 54.52% and higher-order only 3.92%, versus a uniform distribution (30.39%/38.24%/31.37%) in traditional learning.

"In the agent-enriched learning contexts, cognitive engagement exhibits a non-uniform distribution that differs from traditional learning models (Lower order = 54.52%, Medium order = 41.56%, and Higher order = 3.92%)."

Discussion¶

Related Claims¶

• Learners in agent interaction became locked in lower-order self-transition loops (e.g., Understand z = 49.08, Application z = 51.48) and rarely transitioned spontaneously to higher-order cognition — a narrower finding that bears on this claim
• University examination questions concentrate at the lower cognitive levels of Bloom's Taxonomy, dominated by comprehension and knowledge — related
• Effective AI help-seeking is itself a skill unevenly distributed among students, and access to an assistant is not equivalent to effective help-seeking — related
• Students' stated pre-interaction intentions only weakly correspond to their enacted help-seeking with an LLM — related
• GenAI-assisted education produces significantly higher higher-order thinking abilities than non-GenAI approaches (g = 0.72) — reports the opposite