CogGen: A Learner-Centered Generative AI Architecture for Intelligent Tutoring with Programming Video
Project Overview
The document explores CogGen, an innovative AI architecture aimed at improving programming education through a personalized, interactive video-based tutoring system. By leveraging generative AI and student modeling, CogGen aligns with the Cognitive Apprenticeship framework to deliver customized learning experiences. It features advanced components such as video segmentation, a conversational tutoring engine, and a dynamic student model that adapts instructional content according to individual progress. Evaluations of CogGen indicate its effectiveness in segmenting educational videos and generating tailored instructional materials, successfully addressing common challenges in programming education. Overall, the findings highlight the potential of generative AI to enhance engagement and learning outcomes in educational contexts, particularly in complex subjects like programming.
Key Applications
CogGen
Context: Video-based programming education for students learning programming concepts and coding skills.
Implementation: CogGen segments programming videos into learning goals and uses a conversational tutoring engine to deliver personalized instruction based on a student model.
Outcomes: Achieved 76.9% segmentation accuracy, strong pedagogical alignment, personalized learning experiences, and effective instructional guidance.
Challenges: Requires videos to have a modular structure with discrete learning goals; effectiveness across diverse programming tutorials is yet to be fully evaluated.
Implementation Barriers
Technical Barrier
Challenges in maintaining focus and coherence in multi-turn conversations with large language models, as well as their tendency to provide verbose responses.
Proposed Solutions: Implementing a structured approach to prompt generation to ensure controlled and relevant message generation tailored to specific learning needs.
Structural Barrier
The assumption that programming videos have a modular structure that aligns with discrete learning goals.
Proposed Solutions: Future versions could incorporate adaptive segmentation using temporal coherence and multimodal cues to better align with instructional goals.
Project Team
Wengxi Li
Researcher
Roy Pea
Researcher
Nick Haber
Researcher
Hari Subramonyam
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Wengxi Li, Roy Pea, Nick Haber, Hari Subramonyam
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai