Survey on Plagiarism Detection in Large Language Models: The Impact of ChatGPT and Gemini on Academic Integrity
Project Overview
The document examines the influence of generative AI, particularly Large Language Models (LLMs) such as ChatGPT and Gemini, on education, emphasizing their dual role in enhancing learning and posing challenges to academic integrity. It highlights an increase in academic misconduct, primarily through plagiarism facilitated by AI-generated content, creating significant hurdles for educators in detecting such work. The analysis reviews current plagiarism detection tools, noting their limitations in addressing the sophisticated nature of AI outputs. Furthermore, it suggests alternative educational strategies aimed at countering the negative impacts of generative AI on academic honesty while promoting its potential benefits for learning. Overall, the text underscores the necessity for educators to adapt to the evolving landscape of AI in education, balancing the integration of these technologies with the preservation of academic integrity.
Key Applications
AI Tools for Content Generation and Detection
Context: Higher education settings, particularly among college and university students, where AI tools are used to generate essays, solve programming assignments, and assist with standardized tests, while educational institutions utilize detection tools to identify AI-generated content and plagiarism.
Implementation: Students use AI technologies such as ChatGPT and Gemini to complete assignments and exams by generating essays and solving problems. Simultaneously, educational institutions implement tools like Turnitin and other AIGC detection systems to analyze submitted work against large databases to identify both traditional plagiarism and AI-generated content.
Outcomes: Increased instances of academic misconduct with a significant percentage of students admitting to using AI for homework, and varying success in detecting plagiarism. Detection tools show challenges in identifying AI-generated content, leading to false positives and misclassifications.
Challenges: Difficulty in detecting AI-generated work leads to concerns about the loss of creativity and learning ability among students. Furthermore, there are high rates of false positives in detection tools, where human-written text is mistakenly classified as AI-generated, necessitating continuous updates to detection algorithms.
Implementation Barriers
Technical and Ethical
Current plagiarism detection tools struggle to accurately identify AI-generated text. Additionally, there are concerns about academic integrity and the misuse of AI tools for cheating.
Proposed Solutions: Development of more sophisticated algorithms that can differentiate between AI-generated and human-written content, alongside the implementation of educational policies that promote ethical use of AI in academia and enhance awareness about plagiarism.
Project Team
Shushanta Pudasaini
Researcher
Luis Miralles-Pechuán
Researcher
David Lillis
Researcher
Marisa Llorens Salvador
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Shushanta Pudasaini, Luis Miralles-Pechuán, David Lillis, Marisa Llorens Salvador
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