Leveraging Explainable AI for LLM Text Attribution: Differentiating Human-Written and Multiple LLMs-Generated Text
Project Overview
The document examines the role of generative AI, especially large language models (LLMs), in education, addressing both its potential benefits and risks. It underscores concerns regarding plagiarism and the potential degradation of students' writing skills as AI-generated content becomes more prevalent. A key study discussed in the document investigates the application of machine learning (ML) and explainable AI (XAI) techniques to effectively attribute text to its human or AI origins. This research emphasizes the crucial need for academic integrity and the development of robust detection tools to identify AI-generated material. Ultimately, the document advocates for a balanced approach to integrating generative AI in educational settings, aiming to leverage its capabilities while safeguarding the essential skills and ethical standards of students.
Key Applications
ML and XAI techniques for text attribution
Context: Educational settings, target audience includes educators and students concerned about plagiarism.
Implementation: Utilized ML algorithms like Random Forest and Recurrent Neural Networks to classify text as human-written or generated by LLMs.
Outcomes: Achieved high accuracy (up to 98.5%) in distinguishing between human and LLM-generated text, surpassing existing tools like GPTZero.
Challenges: Difficulty in accurately classifying texts from multiple LLMs due to similarities in writing styles.
Implementation Barriers
Ethical
The reliance on AI tools by students may deteriorate their writing skills and compromise academic integrity.
Proposed Solutions: Implement educational policies and tools that promote ethical use of AI, alongside teaching students essential writing skills.
Technological
Existing AI detection tools like GPTZero have limitations, including false positives and inability to recognize some LLM-generated texts.
Proposed Solutions: Develop more advanced models that leverage ML and XAI techniques for better accuracy and reliability in text attribution.
Project Team
Ayat Najjar
Researcher
Huthaifa I. Ashqar
Researcher
Omar Darwish
Researcher
Eman Hammad
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Ayat Najjar, Huthaifa I. Ashqar, Omar Darwish, Eman Hammad
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