NoHands: Non-Human/AI Neoliterature Detection System 

This project was done as a project for my AI technical elective course in my 4th year.

Design problem: Use the AI techniques learnt during the course to solve a real-life problem. Techniques learnt include: game theory, machine learning, AI search techniques, problem evaluation techniques, performance metric evaluation, generative AI, NLPs, and adversarial games

Solution: This project focuses on the development and evaluation of a deep learning model for text classification as either human-written and AI-generated. Two solution architectures are implemented: the first imports existing word embeddings from a pre-trained model, and the second generates document embeddings from the training data. In addition, both architectures use different neural network structures for classification. Performance is evaluated using standard metrics such as accuracy, precision, recall, and F1-score. Furthermore, qualitative analysis is conducted to examine the model's ability to correctly classify text from both classes from a randomly generated pool of samples. Using Doc2vec vectorization with a vocabulary adapted to the training corpus resulted in a mean F1-score of up to 97%. This project aims to contribute to the development of effective text classification models for distinguishing between human and AI-generated text, with potential applications in various domains such as academia and fake news detection. 

Role: My responsibilities included topic selection, scope narrowing, choosing the appropriate AI techniques to implement, periodic documentation, and writing a Python program to train, test, and validate our neural network design. This project was completed in a team of 3 students.