Human baseline
Construct an independent Minimum Spanning Tree solution before seeing any AI output.
SAGE mapping · GenerateSAGE in practice · international implementation
Critical AI literacy in a UK–China computing classroom.
At Oxford Brookes College, Chengdu University of Technology, SAGE became a compact classroom intervention. Students solved an algorithmic problem themselves, challenged a plausible but flawed AI answer, justified corrections and tested their reasoning under a new constraint.
131undergraduate students completed the activity and survey
2.84 → 3.63mean MST confidence before and after, on a 1–4 scale
dz = 0.91large paired-samples effect size, p < .001
57.3%identified an AI error and supported it with valid MST reasoning
01 / Intervention design
Human reasoning first.
AI as an object of critique.
The five-stage workflow translated SAGE into a normal tutorial. Its design moved students from independent problem solving to comparison, correction, contextual adaptation and reflection.
AI solution review
Inspect a plausible AI response containing several non-optimal edge selections.
SAGE mapping · EvaluateAI solution evaluation
Identify errors, explain the violated MST principles and propose better edges.
SAGE mapping · RefineSynthesis under constraint
Rework the network when one location becomes highly congested and should be avoided.
SAGE mapping · AuditReflection
Explain responsible AI use and the need to verify technical answers.
SAGE mapping · ReflectDefend was deliberately out of scope. This was a formative, non-graded tutorial, so the supervised assurance step was not implemented. The paper proposes a graded follow-up that adds Defend and tests students’ ownership of their reasoning under supervised conditions.
02 / The learning mechanism
A wrong answer became
a learning resource.
The intervention did not ask students to improve a blank AI draft. It gave them prior ownership of a solution, then made AI’s reasoning available for disciplined challenge.
Minimum Spanning Tree reasoning made verification visible.
Optimisation problems suit critical AI literacy because errors can be located, compared and defended. Students could not simply say that the AI was wrong; they had to connect a correction to minimum-edge selection and total-distance reasoning.
The added congestion constraint then tested whether students could adapt the solution rather than repeat a memorised procedure.
- 01Find the faulty choiceIdentify a genuine non-optimal edge in the generated solution.
- 02Justify the errorExplain why the selection conflicts with Minimum Spanning Tree principles.
- 03Replace it with evidencePropose a shorter valid edge and explain why the corrected network is preferable.
- 04Transfer the reasoningAdapt the network under a contextual constraint and reflect on when AI output should be trusted.
03 / What the study found
Promising evidence,
with clear limits.
The results combine a significant pre–post confidence change, survey perceptions, an open-ended reasoning check and instructor observations. They support feasibility and early transfer, while leaving stronger causal and longitudinal questions open.
A measurable shift in one session.
Mean MST confidence rose by 0.79 points, from 2.84 to 3.63 on the study’s 1–4 scale. The paired comparison was statistically significant, t(130) = 10.36, p < .001, with a large effect size.
Students became more cautious about AI output.
The strongest reflection result was being less likely to accept an AI-generated technical solution without checking it: 3.57 out of 4. Comparing solutions also helped students identify mistakes.
Critique was visible—and incomplete.
75 of 131 responses correctly identified and justified an AI error. The remaining responses show why confidence alone is insufficient and why repeated practice in disciplinary justification matters.
Interpret with care.
The intervention took place at one institution, in one tutorial activity, without a control group. Confidence and perceptions were self-reported, the open-ended responses were brief, and the study did not test long-term performance. Instructor observations provided context rather than a formal multi-observer qualitative dataset.
04 / Read the research
The accepted paper is available directly.
The conference paper reports the research questions, tutorial design, participant profile, statistical analysis, student response patterns, classroom observations, limitations and proposed next study.
Suggested in-press citationFakirah, M., Elkhodr, M., & Hussain, S. (2026). “Fostering Critical AI Literacy in UK–China Transnational Computing Education: An Implementation of the SAGE Framework.” In Proceedings of the 2026 2nd International Conference on Artificial Intelligence and Educational Systems (ICAIES 2026), Chengdu, China, 24–26 July 2026. In press.