Distinguishing AI Art: Authenticity and Expertise

Abstract

The spread of artificial intelligence (AI) capable of generating complex visual art presents new questions. As AI tools, drawing from advancements like latent diffusion models [1], become good at mimicking artistic styles [2], understanding human perceptual responses is important, not only for evaluating the technology itself but also for comprehending its societal impact. This research explores the human ability to distinguish between original artwork and AI-generated versions. The main goals are to measure discrimination accuracy, evaluate judgment confidence, and identify the perceptual cues individuals use. A key focus is how prior artistic expertise influences these abilities when people are faced with advanced AI techniques that transfer artistic styles, such as model fine-tuning or image-prompted adaptation [3].

The proposed investigation focuses on a visual discrimination task. Participants will view a collection of images presented in a consistent artistic style. Within this collection, half of the images will be original works by a contemporary artist, while the other half will be AI-generated versions created to mimic the artist’s style using the aforementioned style transfer techniques. Participants, divided into groups based on their art and design experience, will judge the origin of each image (human or AI). They will also provide insights into their decision-making process, allowing for an exploration of cognitive strategies and specific visual cues.

The analysis will explore patterns in discrimination accuracy and confidence levels across different participant groups, examining if and how expertise affects performance. Qualitative insights into perceptual strategies will also be sought from participant feedback. The research is expected to contribute to evaluating the current level of difficulty of distinguishing AI from original images, as well as the overall effectiveness of accessible AI style transfer tools. This information is valuable for understanding their present potential and inherent limitations, which is relevant for both developers and users of these technologies. Furthermore, the study aims to clarify how artistic expertise aids in detecting AI-generated content and to help understand the cognitive processes behind judgments about art and authenticity in the age of AI. These findings could contribute to our understanding of human-AI interaction in creative domains. Given AI's growing ability to generate stylistically coherent art, understanding how effectively humans can identify its origin is a key area of interest. This research aims to further this understanding by examining the factors that influence our ability to discern AI-created art from human originals, informing both theoretical understanding and practical considerations within the broader discussion on the changing relationship between human creativity and artificial intelligence.

References

[1] R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High-Resolution Image Synthesis with Latent Diffusion Models,” 2022. [Online]. Available: https://arxiv.org/abs/2112.10752.

[2] A. Y. J. Ha, J. Passananti, R. Bhaskar, S. Shan, R. Southen, H. Zheng, and B. Y. Zhao, “Organic or Diffused: Can We Distinguish Human Art from AI-Generated Images?” 2024. [Online]. Available: https://arxiv.org/abs/2402.03214.

[3] H. Ye, J. Zhang, S. Liu, X. Han, and W. Yang, “IP-Adapter: Text Compatible Image Prompt Adapter for Text-to-Image Diffusion Models,” 2023. [Online]. Available: https://arxiv.org/abs/2308.06721.