Detecting Deepfakes and Artificial Intelligence-Generated Art

Recent years have seen innovative advancements in Artificial Intelligence (AI) in the realms of image and video processing, speech and audio recognition, and pattern recognition, leading to the boom in AI art and deepfakes. The evolution of ChatGPT, Dalle-2, and Midjourney has sparked debate about whether AI art qualifies as true art and whether it will eventually displace human artists and creators. AI art generation models are trained on large datasets, often consisting of copyrighted art. Due to this, the resulting picture produced is frequently a manipulated version of the art piece, namely through various techniques - CutMix, Adversarial Data Poisoning, Inpainting, and Style Transfer. This manipulation constitutes a violation of the rights held by the artists. On a similar note, deep fakes pose threats to privacy and security. Society is entering a new age of misinformation and confusion, where they are victims of deepfake phishing scams, identity theft, and malicious political interference. These issues necessitate the development of state-of-the-art algorithms to detect real-life photos and videos, as well as painting and art manipulation. This paper dives deep into the topic of AI art and Deepfake detection, highlighting the most efficient machine learning classification techniques on the DeepfakeArt Challenge and Deepfake Detection Challenge dataset. It aims to highlight the best algorithm that provides reliable results and assists future research in this newly emerging area.