Generative AI Detection

Overview

This project is a web application built with ReactJS (frontend) and Python (backend) that detects whether a given text is human-written or AI-generated. It participates in the Voight-Kampff Generative AI Authorship Verification 2024 challenge.

Table of Contents

• Generative AI Detection
  • Overview
  • Table of Contents
  • Features
  • Modules and Libraries
    • Flask
    • Flask-CORS
    • PyTorch
    • Transformers (Hugging Face)
    • Regular Expressions (re)
    • OrderedDict (collections)
  • Algorithm and Use Case
    • Perplexity
    • Burstiness
    • Thresholding
    • API Usage
    • Use Cases
  • Summary:
  • Installation
  • Usage
  • Data
  • Evaluation Metrics
  • Key Changes:
  • Contributing
  • License

Features

• AI vs Human Detection: Classifies text as human-written or AI-generated.
• Machine Learning Models: Uses various models for accurate classification.
• ReactJS Frontend: Provides a user-friendly interface.
• Python Backend: Handles AI detection logic with robust backend processing.

Modules and Libraries

Flask

• Lightweight Python web framework for building web applications.
• Provides HTTP request routing and response generation.

Flask-CORS

• Extension to enable Cross-Origin Resource Sharing (CORS).
• Allows server to respond to requests from different origins.

PyTorch

• Deep learning library used to load and run models.
• Runs the GPT-2 model for computing perplexity.

Transformers (Hugging Face)

• Uses GPT2LMHeadModel and GPT2TokenizerFast for loading the GPT-2 model and tokenizer.
• GPT-2 is a pre-trained language model for text generation and analysis.

Regular Expressions (re)

• Performs text processing tasks like character extraction and sentence splitting.

OrderedDict (collections)

• Maintains order of dictionary entries for structured results.

Algorithm and Use Case

The core algorithm uses GPT-2 to calculate Perplexity and Burstiness of input text:

Perplexity

• Measures how well the model predicts a sentence.
• Lower perplexity implies human-written text; higher implies AI-generated.
• Computed using the negative log likelihood of each word.

Burstiness

• Measures variation in perplexity across lines.
• High burstiness often indicates AI-generated content.

Thresholding

• Perplexity < 60: Likely AI-generated.
• Perplexity 60–80: Most likely AI, but requires more text for confirmation.
• Perplexity > 80: Likely human-written.

API Usage

• Provides a POST route (/) accepting JSON payloads with a text field.
• Returns label ("AI-generated" or "Human-written") with perplexity and burstiness scores.

Use Cases

• Content authenticity checks (articles, blogs, essays)
• AI detection in education (preventing AI plagiarism)
• Content moderation on social media or forums

Summary:

Modules used: Flask, Flask-CORS, PyTorch, transformers, regex, and OrderedDict.
Algorithms used: Perplexity (text predictability), Burstiness (variation in sentence predictability), and thresholding for labeling the text as AI or human-written.
Use case: AI vs. human text detection, content authenticity verification.

Installation

To set up the project locally, follow these steps:

1. Clone the repository:

   git clone https://github.com/ankitsharma-tech/Generative-AI-Detection.git
   cd Generative-AI-Detection

2. Install frontend dependencies:

   cd frontend
   npm install

3. Install backend dependencies:

   cd ../server
   pip install -r requirements.txt

4. Start the backend server:

   python app.py

5. Start the frontend server:

   npm start

Usage

1. Open your browser and go to http://localhost:3000.
2. Upload a pair of texts (one human-written and one AI-generated).
3. Click on the "Analyse" button to see the results.

Data

• The dataset used in this project consists of a collection of human-written and AI-generated texts.
• Texts are analyzed to calculate metrics like Perplexity and Burstiness to determine their likely origin (AI or human).

Evaluation Metrics

• Perplexity: Measures how well the AI model predicts the next word in a text. Lower perplexity suggests human authorship, while higher perplexity suggests AI generation.
• Burstiness: Measures the variation in perplexity across different lines of text. Higher burstiness often indicates AI-generated text.

Key Changes:

1. Reorganized the content under appropriate headings.
2. Added a Data section for clarity on the dataset.
3. Reformatting of steps in Installation to improve clarity.

Contributing

Contributions are welcome! If you have suggestions for improvements or new features, please fork the repository and submit a pull request.

License

This project is licensed under the MIT License. See the LICENSE file for details.