Desirable Characteristics for AI Teaching Assistants in Programming Education
Project Overview
The document explores the integration of generative AI in education through the implementation of a digital teaching assistant called CodeHelp, which utilizes a large language model (LLM) in an introductory programming course. It addresses the challenges that human teaching assistants encounter and illustrates how AI can enhance the learning experience by offering timely and personalized feedback to students. The findings indicate that students favor scaffolding that encourages their independence in problem-solving rather than simply providing direct answers. Through the collection of student feedback, the study identifies key characteristics and features that students deem essential for an effective digital teaching assistant, emphasizing the need for tools that foster meaningful learning experiences. Overall, the document highlights the potential of AI in supporting educational processes while addressing students' preferences for engagement and autonomy in their learning journeys.
Key Applications
CodeHelp - an LLM-powered digital assistant for programming courses
Context: Introduced in an introductory programming course at the University of Auckland, targeting undergraduate students.
Implementation: CodeHelp was deployed in a structured lab activity where students interacted with it for debugging and code writing tasks. Students provided feedback on its effectiveness.
Outcomes: Students valued the tool for its ability to provide instant support and preferred scaffolding that enables them to solve problems independently.
Challenges: Students may become overly dependent on the tool if the guardrails are too constraining, and there are concerns about its ability to promote meaningful learning experiences.
Implementation Barriers
Dependence on Technology
Students may become overly reliant on the digital TA for solutions rather than developing their own problem-solving skills, raising concerns about the effectiveness of AI-generated feedback compared to that from human tutors.
Proposed Solutions: Implement guardrails in the digital TA to encourage students to engage with the learning process and retain autonomy, and ensure that the digital TA is designed to provide accurate, clear, and context-appropriate feedback.
Project Team
Paul Denny
Researcher
Stephen MacNeil
Researcher
Jaromir Savelka
Researcher
Leo Porter
Researcher
Andrew Luxton-Reilly
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Paul Denny, Stephen MacNeil, Jaromir Savelka, Leo Porter, Andrew Luxton-Reilly
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