Human or Algorithm? The Visual Turing Test of AI-Generated Images

With the advancement of artificial intelligence (AI) technology, the application of AI in generating digital art images has become increasingly common. However, whether people can distinguish images created by the latest AI painting tools from those created by humans, as well as the strategies and success rate of such differentiation, remains to be explored. This study employs a double-blind experimental method, combining a visual Turing test and in-depth interviews, to investigate participants’ ability to distinguish between human-created and AI-generated images, the strategies they use, and the success rate of their distinctions. The results show that participants’ average accuracy in recognizing AI-generated images was 61.67%, higher than the traditional Turing test benchmark of 30%, but 38.33% of participants still failed to accurately distinguish between the images. Participants primarily use three strategies to differentiate the images: details and logic, aesthetic experience, and Human-like characteristics and material properties, with recognition success rates of 75.7%, 73.05%, and 64.5%, respectively. This study reveals the potential of AI in the field of visual arts while also highlighting the advantages of human observation and logical reasoning. It provides empirical evidence for future AI art creation and recognition.