Explainable AI Approach for detecting Generative AI Imagery

Abstract

The rapid advancement of Artificial Intelligence (AI) and machine learning, particularly deep learning models such as Convolutional Neural Networks (CNNs), has revolutionized image classification across diverse fields, including healthcare, autonomous vehicles, and digital forensics. However, the proliferation of AI-generated images, commonly referred to as deepfakes, has introduced significant ethical, societal, and security challenges. Deepfakes leverage AI to create highly realistic yet synthetic media, complicating the ability to differentiate between authentic and manipulated content. This has heightened the need for robust tools capable of accurately detecting and classifying such media to combat the risks of misinformation, fraud, and erosion of public trust.

Traditional models, while effective in classification, often lack transparency in their decision-making processes, limiting stakeholder trust. To address this limitation, this study explores the integration of Explainable AI (XAI) techniques, such as SHAP (SHapley Additive exPlanations), with CNNs to enhance interpretability and trust in model predictions. By employing CNNs for high-accuracy classification and XAI methods for feature-level explanations, the research aims to contribute to digital forensics and content moderation, offering both technical reliability and transparency. This study highlights the critical need for trustworthy AI systems in the fight against manipulated media, providing a framework that balances efficacy, transparency, and ethical considerations.