Presentation

· Contributors · Organizations · Search Program · Flagged · Happening Now

Rolling the Dice: Examining Human Ability to Detect AI-generated Fake Images in e-Commerce Context

DescriptionImage-generating AI applications were intended to enable users to edit existing photos, apply artistic styles to otherwise mundane photos, create social media photos, and design flyers. However, there have been growing concerns about unethical use of AI-generated images. As the first step toward the resolution of these potential threats, the current study aims to examine human ability to distinguish AI-generated fake images that can be potentially exploited in e-commerce. A 2*2 mixed factorial design was used. Between-subject variable was veridicality of the fake images, or the degree to which the images look real, with two levels: high veridicality (HV) and low veridicality (LV). Within-subject variable was time pressure during image screening tasks with two levels: high time pressure (HTP) and low time pressure (LTP). Our study found that the detection accuracy of participants in both HTP conditions did not significantly different from a random chance, or 20% (=1/5). On the contrary, participants in both LTP conditions showed significantly higher detection accuracy. Especially, participants in LTP condition who screened LV images showed the greatest difference from the random chance. Also, results indicated that there was a significant main effect of veridicality and time pressure on detection accuracy.

Contributors