Skip to main content Skip to navigation

Reforming Physics Exams Using Openly Accessible Large Isomorphic Problem Banks created with the assistance of Generative AI: an Explorative Study

Project Overview

The document outlines the transformative role of generative AI, particularly large language models like GPT-3, in reforming assessment practices within STEM education, specifically in physics. It emphasizes the creation of extensive isomorphic problem banks that offer a diverse range of assessment problems aimed at reducing rote memorization while ensuring fairness and appropriate difficulty levels. Exploratory studies indicated that when students were granted open access to these AI-generated problem banks, there was no significant change in exam difficulty or an increase in rote learning behaviors. These findings suggest that the integration of generative AI in education could enhance assessment methods, fostering a more effective learning environment that promotes critical thinking and problem-solving skills among students.

Key Applications

Open Isomorphic Problem Banks

Context: Introductory physics course at the University of Central Florida with 328 registered students, including diverse demographics.

Implementation: Large numbers of isomorphic physics problems were created with the assistance of GPT-3 and other tools. Students had access to these problem banks as practice material prior to mid-term exams.

Outcomes: Results indicated that using open isomorphic problem banks did not significantly change exam difficulty or promote rote memorization. Students performed comparably on open bank and transfer problems.

Challenges: Potential challenges included maintaining the quality and difficulty of generated problems, students may still rely on rote memorization, and ensuring equitable access to problem context.

Implementation Barriers

Technical Barrier

Maintaining the quality and difficulty of generated isomorphic problems. Involvement of human experts in the process to evaluate the clarity and appropriateness of problems.

Proposed Solutions: Human experts should be involved to ensure the quality and appropriateness of generated problems.

Equity Barrier

Certain isomorphic problem variations may not be considered isomorphic by all cultural backgrounds, potentially leading to bias. Future studies should focus on ensuring that the generated problems are culturally inclusive and do not reinforce existing biases.

Proposed Solutions: Future studies should prioritize cultural inclusivity in generated problems to avoid reinforcing existing biases.

Project Team

Zhongzhou Chen

Researcher

Emily Frederick

Researcher

Colleen Cui

Researcher

Munaimah Khan

Researcher

Christopher Klatt

Researcher

Mercedith Huang

Researcher

Shiyang Su

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Zhongzhou Chen, Emily Frederick, Colleen Cui, Munaimah Khan, Christopher Klatt, Mercedith Huang, Shiyang Su

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

Let us know you agree to cookies