A Bidirectional LSTM Deep Neural Network Model With SSA Based Optimizer to Detect Human-Written and AI-Generated Content in Arabic Language

Advancements in AI technologies like ChatGPT are very much equipped and capable of writing articles within a short span of time. Use of artificial intelligence in education needs to be well organized and needs to go beyond technical accuracy, this opens a door to various unethical applications across multiple domains. Students rely on these AI tools, instead of coming up with their own idea and leads to reduce the student’s capacity to analyze information and to form a judgment. In other sectors like medical as well as in the field of scientific research there is an increase in use of ChatGPT for crafting abstracts. Due to these reasons, there is need to focus on the identification of human text and ChatGPT-generated text. To the best of our knowledge, a limited AI detector has been trained on human-written Text and AI-generated Text from ChatGPT in Arabic. Our study reveals that there is a lack of AI detectors to identify Arabic human-written text. The current work uses word embedding and count vectorizers for the feature extraction process, a Sparrow Search Algorithm for parameters tuning with bidirectional Long Short-Term Memory Recurrent Neural Network (SSA-LSTM model) model designed specifically for Arabic language to detect both human as well as AI Generated text. The performance of the proposed SSA-LSTM technique was investigated on benchmark databases and the outcome demonstrated the advantage of the SSA-LSTM system over other recent methods with a maximum accuracy of 97.17 % with lower complexity.