Detecting AI-Generated Text Using Machine Learning and Deep Learning Approaches

Abstract

Recent advancements in natural language
processing have the potential to enable artificial
intelligence systems to produce text indistinguishable
from human-authored content. Such developments
could lead to significant ethical, legal, and societal
implications. This study addresses this challenge by
designing a robust AI detection model to differentiate
between AI-generated and human-written text. To
achieve this, we employ k-fold cross-validation to evalu
ate a range of established machine learning and deep
learning models, including Logistic Regression, Extra
Trees Classifier, CNN, RNN, and LSTM networks. Our
experimental results reveal that the CNN-based model
outperforms other approaches in accurately identifying
AI-generated content. In addition to presenting our
f
indings, we thoroughly review existing research in
AI-generated text detection, comprehensively analyzing
current methodologies and their limitations. Our testing
demonstrates promising outcomes, with the proposed
CNN-based approach emerging as the most effective
solution.
Specifically, the LSTM and RNN models
achieve accuracies of 0.83, while the Detect CNN model
attains the highest accuracy of 0.85. Beyond technical
performance, we also explore the broader societal
implications of this research, emphasizing its potential
benefits across various sectors. Furthermore, we
address critical ethical considerations and environmental
sustainability concerns, underscoring the need for
responsible development and deployment of such
technologies.