Opening a New Educational AI Paradigm

Building a truly intelligent educational ecosystem centered on learners, transcending the limitations of existing AI tutors.

Limitations of Current Approaches

  • ●Focus solely on algorithmic performance, failing in real educational interactions.
  • ●Inadequate response to unpredictable questions, generating only generic feedback.
  • ●Lack of deep understanding of the ultimate beneficiaries of technology - the 'learners'.

Paradigm Shift in This Research

  • ●Shifting perspective from Tutor-Centric → Learner-Centric.
  • ●Verifying educational effectiveness using 'high-fidelity student agents'.
  • ●Accelerating technological advancement by shortening the hypothesis-experiment-verification cycle from months to days.

Organic Collaboration Model: Consortium

Building a world-class AI education platform centered on four axes that maximize each institution's core capabilities.

Cheju Halla University

Student Agent Modeling

Laying the foundation for research by providing realistic virtual learner environments.

Saltlux

AI Engine & Platform

Developing powerful AI engines and platforms that serve as the brain of the tutor.

Yonsei University

LLM Tutor Agent

Interacting with virtual learners and advancing educational strategies.

KAIST

Evaluation & Verification

Precisely measuring and providing feedback on the educational effectiveness of interactions.

Core Technologies by Institution

Achieving project goals through detailed research plans and core technologies that each institution specializes in.

Implementation of 'Living' Virtual Learners

Developing dynamic digital twins that simulate real learners' behavior, cognition, and emotions in real-time.

  • Knowledge Tracing (KT) and Explainable KT (XKT): Enhancing Transformer models and visualizing reasoning evidence through SHAP, GRAD-CAM, etc., to ensure transparency.
  • LLM-based Virtual Student Agents: Sophisticated modeling of various personas and misconception patterns such as 'overconfident students' and 'hesitant students'.
  • Multimodal Learning Analytics (MMLA): Integrating gaze, biometric signals, etc., to infer real-time states like confusion, boredom, and distraction, and reflect them in agent behavior.

Integrated Research Flow Centered on Student Agents

Beyond simple role division, maximizing the completeness of research outcomes through organic feedback loops.

Step 1: Simulation Environment Provision

Step 2: Development and Evaluation Environment Integration

Step 3: Integrated Results Analysis and Feedback

Step 4: Data-driven Simultaneous Model Improvement

Details

Click on each step to see detailed explanations.

AI Tutor Strategy Simulator

Simulating expected effects based on various student personas and AI tutor questioning strategies.

Expected Impact

Technological, industrial, national, and social changes that this research will bring.

(Technology) Innovative Research Environment

Testing numerous hypotheses safely, quickly, and at low cost through 'complete digital twins' of educational environments.

(Industry) EdTech Ecosystem Contribution

Providing 'standardized verification environments' to shorten technology development cycles for startups and reduce market entry barriers.

(National) Global Competitiveness

Leading global performance verification standards in the field of 'trustworthy educational AI'.

(Social) Building Trust in Educational Settings

Increasing social acceptance of AI technology by presenting data-based concrete evidence.