Review of Detecting Text generated by ChatGPT Using Machine and Deep-Learning Models: A Tools and Methods Analysis
Recently, generative models, such as ChatGPT, have gained considerable attention because of their capacity to generate text almost identical to that produced by humans. However, ChatGPT raises several concerns, particularly regarding the integrity of academic work, the protection of personal inform...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
University of Diyala
2025-03-01
|
| Series: | Diyala Journal of Engineering Sciences |
| Subjects: | |
| Online Access: | https://djes.info/index.php/djes/article/view/1676 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Recently, generative models, such as ChatGPT, have gained considerable attention because of their capacity to generate text almost identical to that produced by humans. However, ChatGPT raises several concerns, particularly regarding the integrity of academic work, the protection of personal information and security, the reliance on artificial intelligence (AI), the evaluation of learning, and the precision of information. Distinguishing between writing generated by machines and text that humans wrote is one of the most critical issues at present. The purpose of this literature review is to provide a comprehensive, up-to-date analysis of the most recent methods for identifying text that ChatGPT created. It examines more than 60 academic papers, especially research articles published after the model’s release in 2022, and analyzes state-of-the-art machine learning, deep learning, and hybrid approaches for detecting AI-generated text. The review categorizes detection methods into statistical models, transformer-based architectures, perplexity-based techniques, and human-assisted evaluation. The findings indicate that deep learning models, particularly the Robustly Optimized BERT Pretraining Approach (RoBERTa) and Cross-lingual Language Model with RoBERTa Architecture, have high detection accuracy (up to 99%), whereas traditional statistical methods exhibit limitations in distinguishing complex AI-generated content. This work recommends the use of machine and deep learning techniques and human reviewers in ongoing efforts to distinguish between AI-generated and human-written text. However, given the increasing sophistication and complexity of models, such as ChatGPT, detection techniques have to be continuously improved and innovated to ensure reliability and maintain the integrity of content across various sectors.
|
|---|---|
| ISSN: | 1999-8716 2616-6909 |