The Effect of Inductive Learning on Differentiating Al-Generated from Human-Made Photographs

Mutlu, Timucin (2025) The Effect of Inductive Learning on Differentiating Al-Generated from Human-Made Photographs. Bachelor thesis, Psychology.

Text S5167159Bachelor-ThesisTimucin-Mutlu.pdf Restricted to Repository staff only Download (951kB)

Abstract

The rapid advancement of generative artificial intelligence has made AI-generated images increasingly difficult to distinguish from human-made photographs, raising concerns about media authenticity. This study investigated whether humans can learn to identify AI- generated versus human-made photographs through inductive learning training. Further, the role of image content and gender on detection accuracy and learning was also examined. A total of 193 participants (137 females, 56 males) were randomly assigned to either a training condition (n=91) or a control condition (n=102). The training group viewed 78 labeled images (AI-generated or human-made) across three content categories: landscapes, everyday humans, and artistic humans. Both groups completed a testing phase with 42 unlabeled images across the same categories. Results revealed a complex pattern of learning effects. While overall accuracy remained identical between groups (57.2%), training significantly improved AI detection accuracy at the cost of reduced accuracy for identifying human-made images. At baseline, participants showed higher accuracy for AI-generated everyday human images than landscapes, but performed worse on artistic human images. Additionally, women demonstrated significantly higher accuracy than men at identifying AI-generated artistic human images. Training benefits did not differ significantly between image categories, and no substantial gender differences in learning capacity were observed. The results suggest that while training may enhance humans’ ability to spot AI-generated images, it could also introduce a bias, causing them to label more real images as artificial. These results highlight the challenges facing society as AI-generated visual content becomes increasingly sophisticated and difficult to detect.

Item Type:	Thesis (Bachelor)
Supervisor name:	Gutzkow, B.
Degree programme:	Psychology
Differentiation route:	None [Bachelor Psychology]
Date Deposited:	10 Jul 2025 07:37
Last Modified:	10 Jul 2025 07:37
URI:	http://gmwpublic.studenttheses.ub.rug.nl/id/eprint/5431

Actions (login required)

View Item