Detection and classification of ChatGPT-generated content using deep transformer models.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Machine Learning

Background:

• Rapid advancements in AI, especially neural networks and large language models (LLMs) like ChatGPT, have enabled sophisticated text generation.
• The increasing prevalence of AI-generated text raises concerns regarding academic integrity, misinformation, and privacy.
• Existing methods for detecting AI-generated content often show questionable performance, particularly with advanced LLMs.

Purpose of the Study:

• To develop and evaluate machine learning models for accurately detecting and classifying AI-generated text.
• To establish a robust baseline for differentiating between human-authored and AI-generated content.
• To mitigate potential misuse of AI text generation tools.

Main Methods:

• Compiled a dataset of human-written and AI-generated (ChatGPT) text.
• Trained and evaluated various machine learning and deep learning models, focusing on transformer-based architectures.
• Assessed model efficacy using metrics such as F1-score and accuracy.

Main Results:

• A custom RoBERTa-based deep learning model achieved an F1-score of 0.992 and accuracy of 0.991.
• DistilBERT demonstrated strong performance with an F1-score of 0.998 and accuracy of 0.988.
• These results indicate exceptional capability in detecting AI-generated content.

Conclusions:

• The developed models provide a reliable approach for distinguishing AI-generated text from human writing.
• This research establishes a strong foundation for AI text detection, crucial for addressing academic integrity and misinformation.
• Future work should focus on model generalizability across diverse AI sources and evolving detection challenges.