Home // GPTMB 2024, The First International Conference on Generative Pre-trained Transformer Models and Beyond // View article

Human Perception and Classification of AI-Generated Images: A Pre-Study based on a Sample from the Media Sector in Germany

Authors:
Stephan Böhm

Keywords: Generative AI; AI-based media disruption; AIgenerated images; human perception of AI; identification of AIgenerated images

Abstract:
Recent advances in Generative Artificial Intelligence (AI) have significantly expanded and improved image generation and processing possibilities. Applications, such as DALL-E, Midjourney, and Stable Diffusion have simplified Generative AI for non-technicians and made it accessible to a broad audience. The quality of the generated images has steadily increased in recent months, with photo-realistic representations almost indistinguishable from real photos. AI-based image generation and editing methods are also becoming increasingly accessible for professional use, where high-quality image generation and editing were formerly reserved for specially trained personnel. However, the perception of Generative AI's results and potential depends not only on image quality. Human users may have reservations or a biased assessment of the performance of AI for image generation, for example, because they doubt the creativity of AI or fear the substitution of jobs. Against this background, a pre-study with a sample of N = 172 participants from the media sector in Germany is presented. The participants were asked about their attitudes towards image-generating AI and had to assess a test set of images regarding quality and type of generation. The results show that while minor differences in quality are observed, classification precision is almost independent of the quality rating and the participants' attitudes or experiences. The study supports the conclusion that even representatives from the media sector cannot systematically recognize AI-generated images based on image quality at the current performance level of image-generating Generative AI.

Pages: 15 to 23

Copyright: Copyright (c) IARIA, 2024

Publication date: June 30, 2024

Published in: conference

ISBN: 978-1-68558-182-4

Location: Porto, Portugal

Dates: from June 30, 2024 to July 4, 2024