Navigating the Transition

A System Dynamics Analysis of AI-Augmented Pedagogy�in East Asian Data Science Education

PACIS 2026 | Jakarta, Indonesia | July 8, 2026�Taekyung Kim, Kyung Hee University�Martin Kang, Loyola Marymount University

1. Introduction

Photo: The Korea Economic Daily

Same Lecture,�New Tools?

Generative AI is everywhere, and information is a search away. Yet the classroom still looks the same: one instructor talking, many students listening — a pattern especially hard to break in East Asia's Confucian Heritage Culture.

2. Theoretical Grounding

• AI in Education: From Scaffolding to Substitution

- Existing ITS (Intelligent Tutoring Systems) act as a temporary "scaffold" tailored to the learner's ability

- Today's LLMs substitute for the learner's task itself, depriving students of the "desirable difficulty" of cognitive struggle

• Building the Theoretical Framework

- IS Success Model (DeLone & McLean, 2003): the foundational model for assessing AI outcomes

- Cognitive Offloading (Rosen et al., 2011): skill atrophy that results from delegating cognitive effort to an external agent

- Socio-technical System (Bostrom & Heinen, 1977): the misaligned delay between the technical subsystem (rapid AI adoption) and the social subsystem (slow institutional adaptation of pedagogy)

• System Dynamics (SD)

- Analyzes the "Better-Before-Worse" dynamic — short-term success followed by long-term failure — commonly seen when new technologies are introduced

3. Model Constructs

3. Model Constructs

3. Model Constructs

4. Equations

4. Equations

4. Equations

4. Equations

4. Equations

5. Simulation Calibration

• Core parameters calibrated using large-scale data from the Korea Education and Research Information Service (KERIS), including the RM report series

• Key Calibration Parameters:

- Trust Bias (α = 0.8): higher deference to digital authority than in other cultures (KERIS RM 2020-26)

- Diffusion Speed (β = 1.8): extremely rapid contagion rate of AI-tool spread among students (KERIS RM 2018-27)

- Training Delay (τ = 2.0 years): average lag before pedagogical changes take effect, covering curriculum review, faculty training, and institutional rollout

- Erosion Factor (ε = 0.18): reflects the natural annual decay rate of cognitive skills

- Risk Sensitivity (γ = 30) / Risk Exponent (ρ = 1.5): the critical-hit coefficient for failure to verify AI-generated output

6. System Dynamics Loop

7. Results

7. Results

7. Results

7. Results

8. Discussions

Parasitic Learning Loop

AI robs students of the 'cognitive struggle essential to deep understanding', creating what appears to be rising competence but is actually a ruinous, self-amplifying (Self-reinforcing) structure

Unlike removable 'scaffolding,' it has a parasitic quality: the more it is used, the harder it becomes to stop, steadily eroding critical-thinking ability

Cognitive Offloading

The act of offloading computation the brain should perform onto AI as an external machine. This quantifies how the phenomenon transforms the socio-technical system in a way that is not one-off, but 'cumulative and path-dependent'

Single to Double-loop Learning

Moves away from mindlessly AI copying down answers in single-loop (Single-loop) learning, and calls for a leap — through the school’s pedagogical reform (Training Ramp) — toward 'monitoring and discriminating errors themselves'-driven double-loop (Double-loop) learning

Desirable Difficulties (Bjork 1994)

Forming long-term memory and genuine knowledge necessarily requires 'effortful conditions'. AI anesthetizes this difficulty, creating the illusion of "strikingly fast progress paired with endlessly shallow understanding"

8. Discussions

Synergy of AI-Based Coding within East Asian Education Culture

• Digital Authoritarianism: Due to a cultural tendency toward less skepticism and criticism, students’ 'short-term productivity loop(R1)'operates far faster and more explosively than in Western contexts.
• Training Delay: Because of the hierarchical decision-making structure typical of Korean universities, reforming curricula or teaching methods takes a baseline of 2-3 years 🡪 Students use it,, classes fail to adapt,, and assessment is likely to go wrong.
• Double Asymmetry(the root cause of the crisis): students’ AI adoption is "decentralized and lightning-fast (Fast & Decentralized)" whereas the institution’s adaptation is "centralized and turtle-like (Slow & Centralized).” Instructors, who are responsible for educational quality, are likely to find themselves caught in a painful sandwich between fast-changing students and slow-moving school administration, a potential problem that is difficult to foresee.

Q & A

Thank you for your attention