Learning from Teaching Assistants to Program with Subgoals: Exploring the Potential for AI Teaching Assistants
Project Overview
The document explores the integration of generative AI, specifically AI teaching assistants (TAs) like ChatGPT, within programming education, demonstrating their potential to improve student learning outcomes. By assisting learners in breaking down complex problems into manageable subgoals and generating structured solutions, AI TAs enhance the educational experience. The findings reveal that students utilizing AI TAs not only completed tasks more quickly but also achieved success rates comparable to those who worked with human TAs. Additionally, the document identifies both the advantages and limitations of AI TAs, providing design guidelines to optimize their effectiveness in educational settings. Overall, it presents a promising view of how generative AI can transform the learning process in programming education by combining efficiency with effective support.
Key Applications
Generative AI Teaching Assistants
Context: Introductory programming courses for novice learners.
Implementation: AI TAs were integrated into a learning workflow that includes subgoal formulation, subsolution generation, and solution generation, allowing learners to interact with the AI while solving programming tasks.
Outcomes: Learners using AI TAs solved tasks faster and attempted more tasks, achieving comparable scores to those with human TAs. Positive perceptions of the AI TA were noted, particularly regarding the speed and detail of responses.
Challenges: AI TAs sometimes provided excessive help, which could hinder learning. They also struggled with pedagogical abilities compared to human TAs and sometimes generated inconsistent code.
Implementation Barriers
Pedagogical Limitations
AI TAs may not provide the same level of personalized feedback and understanding as human TAs, potentially weakening the learning experience. Additionally, learners might become too dependent on AI TAs for solutions, which could hinder the development of critical thinking and problem-solving skills.
Proposed Solutions: Implement design guidelines to restrict information provided by AI TAs and teach learners to ask better questions. Encourage learners to reflect on their own solutions and restrict the AI from giving direct answers unless specifically requested.
Communication Challenges
Learners sometimes found it difficult to communicate effectively with the AI, leading to misunderstandings.
Proposed Solutions: Provide templates for effective prompts and encourage practice in formulating questions.
Project Team
Changyoon Lee
Researcher
Junho Myung
Researcher
Jieun Han
Researcher
Jiho Jin
Researcher
Alice Oh
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Changyoon Lee, Junho Myung, Jieun Han, Jiho Jin, Alice Oh
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