Generative AI and the challenge of deepfake detection: a comparative study of AI and human accuracy

The Generative Artificial Intelligence particularly the models such as GANS (Generative Adversial Networks) and EfficientNet has posed several societal and ethical challenges. The rising power of AI and its misuse results in creating hyper-realistic deepfakes. This study aims to explore how well can AI systems and human observers distinguish between real and deepfake images. Not only it brings to light the unethical and moral consequences on public opinions and media trust. A systematic survey of 150 participants from various demographics was performed to find out how much factors like age, gender and familiarity with deepfakes influence the extent of accuracy and the confidence level of the participants in evaluating the factual images. An EfficientNet B0 model was used to test the potential of AI in helping humans detect fake images. This showed that the prediction is generally accurate according to images' categories and participants. There were very high rates of accuracy between AI and human participants who seemed to be overconfident in their assessments. Such behavior indicates the need for better AI tools to reverse the dangers posed by misinformation. Also such studies indicate the necessity of ethical regulations and the fact that the use of AI for deepfake detection is inevitable. The need of the hour is to hold Media literacy campaigns, regulatory policies and future research directions which will help to bring a balance between technology, innovation and public trust.