Towards Mitigating ChatGPT's Negative Impact on Education: Optimizing Question Design through Bloom's Taxonomy
Project Overview
The document explores the impact of generative AI tools, such as ChatGPT, on education, particularly highlighting the challenges they present in assessing student learning. It emphasizes the need for innovative strategies to address these challenges, proposing an evolutionary approach to optimize the use of Bloom’s taxonomy keywords for crafting questions that AI tools find difficult to answer. This strategy is designed to foster critical thinking skills among students and improve the quality of assessments by reducing dependency on AI-generated responses. Through this method, the document aims to enhance educational outcomes by ensuring that assessments remain rigorous and encourage deeper cognitive engagement, ultimately preparing students for a world where AI is increasingly prevalent.
Key Applications
Optimization of Bloom’s taxonomy keywords to generate questions
Context: Educational context in a Data Structures and Representation course; target audience includes students in programming education.
Implementation: An evolutionary algorithm was used to select keywords from Bloom’s taxonomy to create questions that generative AI tools like ChatGPT have low confidence in answering.
Outcomes: The approach reduced ChatGPT’s confidence in answering specific questions from an average of 95% to as low as 0%, thereby promoting better evaluation of student learning.
Challenges: The brute force approach of trying all combinations of keywords was inefficient due to the large search space created by the AI’s ability to restructure questions.
Implementation Barriers
Technical
The brute force method of testing all combinations of Bloom's Taxonomy keywords is computationally intensive and inefficient.
Proposed Solutions: The proposed evolutionary algorithm optimizes keyword selection to effectively reduce AI confidence in responses without exhaustive search.
Cost-related
Limited monthly usage and costs associated with the OpenAI API could restrict scalability and the number of experiments.
Proposed Solutions: Future work aims to mitigate usage limitations and explore cost-effective solutions.
Project Team
Saber Elsayed
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Saber Elsayed
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