Deepfake Detection: Human Performance Versus AI Tools – A Comparison of Accuracy and Effectiveness

Abstract

Image manipulation is a phenomenon much older than digital image handling and generative artificial intelligence (GenAI). In the digital era, researchers have made a distinction between cheapfakes and deepfakes. The creation of cheapfakes requires a relatively low technical editing level and does not depend on any GenAI technology. This study had a focus on deepfakes only and explored the role of GenAI in the generation of deepfakes as well as the role of AI in tools for detecting deepfakes. The term deepfakes refers to high-quality and synthetic media content created with the use of deep learning and generative artificial intelligence. Recent advances in deepfake generation have made such content even more realistic, making it harder to identify. The rise of AI-convincing content is a growing social issue that poses serious challenges for its detection. Although the importance of deepfake detection is widely recognized, research comparing the performance of humans and AI on deepfakes analysis is still in its early stages. This study addresses this gap by conducting a comparison between human analysis and AI-based detection tools to evaluate their accuracy and effectiveness in identifying deepfakes. The testing was based on a set of AI-generated images related to the Israel–Hamas war, which were circulated on social media during 2023-2024. Future research should focus on testing a broader range of AI-based detectors to evaluate their effectiveness against different types of deepfakes, such as video, music, audio, text, including less sophisticated AI-generated manipulations as cheapfakes.