Evidence of a Cognitive Shift in AI Education: How Students Are Rethinking Human Intelligence?

Synthesis: This paper presents a striking longitudinal finding: as AI becomes a routine educational tool, students systematically revalue human intelligence (HI) over artificial intelligence (AI). Drawing on 6 years of classroom poll data (2020–2026) from 471 undergraduate and MSc computer science students, Rekik documents a cognitive shift that progresses through four phases: hype → distrust → trust → dependency.

Overview

Four-phase trajectory of student perception:

1. Hype (2020): Initial excitement slightly favored AI over HI in poll responses
2. Distrust: Emerging skepticism as students encountered AI limitations
3. Trust: Growing reliance on AI tools for coursework
4. Dependency: Habitual use leading to a reappraisal of what makes human intelligence valuable

Converging toward human intelligence. From 2024 onward, a consistent shift toward HI preference emerged across all MSc cohorts. By 2026:

• Technical course (ML/Deep Graph Learning): 65% preferred HI (↑12pp from 2025)
• Design-oriented course (Design Thinking for AI): 90% preferred HI (↑36pp from 2025)

This is a striking reversal from 2020, when AI was slightly favored.

Connections to Faculty Development

For instructors, these results suggest that AI education should explicitly surface the value of human cognition rather than treating AI as a neutral productivity tool. The Faculty Orientations Shape Adoption of AI in Research and Teaching framework shows that faculty AI orientation strongly predicts adoption — this paper adds that student orientation evolves dynamically and may benefit from curricular Scaffolding.

Methodological Notes

• N = 471 students across technical and design-oriented courses
• Poll-based measurement of HI vs. AI valuation
• Longitudinal design spanning 6 years (2020–2026)
• Workshop paper (ICLR HCAIR), not yet peer-reviewed at a major venue

What this means for practice

• Learners. Reassess your own stance on purpose instead of letting it drift with exposure: the reversal toward HI appeared only after sustained use, and the authors' reading of it is a reappraisal of pedagogical framing — courses that teach students to think critically about AI, not just with it, accelerate the shift.
• Learners. Treat AI use as a test of your epistemic agency rather than a productivity default: the authors warn against sliding into dependency and argue that the point of the shift is to use AI for higher abstraction and Metacognition without relinquishing foundational cognitive skills.
• Instructors. Teach critical evaluation about AI inside the course where students use it, not in a separate literacy module: preference for HI climbed 36 percentage-points from 2025 in the design-oriented course (90%) against 12 percentage-points in the technical course (65%), and the paper attributes the sharper swing to how the course framed human intelligence.
• Instructors. Collect an entering-beliefs poll in the opening lecture, before technical instruction, so the course's own starting point is visible — the original protocol used one anonymous forced-choice question at the first session of each course.
• Faculty developers. Plan for the whole trajectory rather than for tool training: the paper frames the four phases (hype, distrust, trust, dependency) as a path to guide learners toward the trust phase while actively preventing a slide into dependency, which implies curricular attention to judgment, synthesis, and moral reasoning under pervasive AI assistance.

Limitations

• The evidence is a single forced-choice question administered to opportunistic classroom cohorts; the authors state it "does not support population-level inference or causal claims."
• The 471 students were polled at the start of AI-focused courses between 2020 and 2026, with sparse observations the authors describe as not population-representative — there is no control condition and no individual-level follow-up.
• The prompt itself is the measurement: a binary machine-intelligence-versus-human-intelligence choice necessarily simplifies a multidimensional relationship and may amplify contrast between the two.
• Cohort composition, institutional context, and contemporaneous social or technological factors may each contribute to the observed pattern, and the four phases are offered as conceptual lenses that may overlap or recur, not as universal sequential stages.

Citation

Rekik, I. (2026). Evidence of a Cognitive Shift in AI Education: How Students Are Rethinking Human Intelligence?. ICLR HCAIR Workshop 2026.