Keystroke Dynamics Against Academic Dishonesty in the Age of LLMs
Project Overview
The document explores the transformative role of generative AI in education, emphasizing both its innovative applications and the challenges it presents. It specifically addresses the issue of academic dishonesty in online learning environments, exacerbated by the accessibility of generative AI tools that can produce written content. To combat this, the authors propose a novel method based on keystroke dynamics, which effectively differentiates between authentic student writing and AI-assisted outputs. The study showcases the method's efficacy across various scenarios, suggesting that it can serve as a valuable complement to existing plagiarism detection systems. Overall, the findings indicate that while generative AI offers significant potential for enhancing educational experiences, it also necessitates the development of new strategies to uphold academic integrity.
Key Applications
Keystroke dynamics-based plagiarism detection
Context: Online education and examinations, targeting students in various disciplines
Implementation: Utilization of keystroke patterns recorded during writing tasks to train a model (TypeNet) that distinguishes between genuine and AI-assisted writing.
Outcomes: Achieved accuracies ranging from 74.98% to 85.72% in condition-specific scenarios, improving detection of academic dishonesty.
Challenges: Variability in typing behavior across different users and contexts, challenges in generalizability of the model across diverse datasets.
Implementation Barriers
Technical Barrier
The model struggles to generalize across different typing behaviors and external factors like keyboard layout and user background.
Proposed Solutions: Future research should focus on developing more robust models that can adapt to these variables, ensuring consistent performance across different contexts.
Ethical Barrier
Potential for high rates of false positives leading to wrongful accusations of plagiarism.
Proposed Solutions: Refinement of detection algorithms to improve accuracy and adaptability across diverse user demographics and academic contexts.
Project Team
Debnath Kundu
Researcher
Atharva Mehta
Researcher
Rajesh Kumar
Researcher
Naman Lal
Researcher
Avinash Anand
Researcher
Apoorv Singh
Researcher
Rajiv Ratn Shah
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Debnath Kundu, Atharva Mehta, Rajesh Kumar, Naman Lal, Avinash Anand, Apoorv Singh, Rajiv Ratn Shah
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai