Towards Detection of AI-Generated Texts and Misinformation

Abstract

Artificial Intelligence (AI) in the form of social text bots has emerged online in social media platforms such as Reddit, Facebook, Twitter, and Instagram. The increased cultural dependency on information and online interaction has given rise to bad actors who use text bots to generate and post texts on these platforms. Using the influence of social media, these actors are able to quickly disseminate misinformation and disinformation to change public perception on controversial political, economic, and social issues. To detect such AI-bot-based misinformation, we build a machine-learning-based algorithm and test it against the popular text generation algorithm, Generative Pre-trained Transformer (GPT), to show its effectiveness for distinguishing between AI-generated and human generated texts. Using a Neural Network with three hidden layers and Small BERT, we achieve a high accuracy performance between 97 % and 99 % depending on the loss function utilized for detection classification. This paper aims to facilitate future research in text bot detection in order to defend against misinformation and explore future research directions.