When XAI Meets CAPTCHA: A Case Study

When XAI Meets CAPTCHA: A Case Study

연구 분야: Verification

학회: 2024 Cyber Awareness and Research Symposium (CARS)

In the rapidly evolving landscape of cybersecurity, the need for robust and transparent defenses against automated attacks is paramount. This paper examines the current landscape of human-centric cybersecurity, focusing on understanding human behavior, cognition, and emotions in relation to cybersecurity practices. We explore how Artificial Intelligence (AI) and machine learning (ML) technologies are being integrated into this framework to create more adaptive and responsive security systems, with a focal point on Completely Automated Public Turing tests to tell Computers and Humans Apart (CAPTCHA). Specifically, we investigate the integration of the eXplainable AI (XAI), using the Local Interpretable Model-Agnostic Explanations (LIME) technique, into CAPTCHA solvers to enhance the transparency of the model performance. By progressively refining a deep learning model designed to solve text-based CAPTCHAs, we demonstrate the impact of increasing training epochs and the number of samples explained by LIME on model accuracy and interpretability. The final results underscore the potential of XAI to improve the transparency and efficacy of ML models in cybersecurity applications, paving the way for more secure and trustworthy digital environments.