Artificial Intelligence (AI) has become increasingly prevalent in society with the rise oftools like Stable Diffusion. This evolution has shown an increase in the use of AI picturesin different fraud cases. To combat this, some researchers are in the process of trainingneural networks to detect AI-generated images. However, these tools are not perfect andare not always commercially available. Thus, to increase resilience against this kind offraud, this study lists practical methods for spotting AI-generated images and measurespeople’s perceived abilities against their actual abilities to determine whether the pictureis generated or not. These results are based on a survey designed to determine standard techniques people use to spot AI images and discern their self-rated ability to detectthem. To conclude, the participants rated their ability to detect generated images lowerthan their actual ability turned out to be. The most effective techniques are looking forincorrect shadows, blurry skin or textures, lack of texture/fine details, incorrect amount ofhands/fingers, abnormal background details, lack of imperfections, poor picture qualityand reflections.
Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:hh-53722 |
Date | January 2024 |
Creators | Eriksson, Arvid, Carlhjelm, Mathias |
Publisher | Högskolan i Halmstad, Akademin för informationsteknologi |
Source Sets | DiVA Archive at Upsalla University |
Language | English |
Detected Language | English |
Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
Format | application/pdf |
Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.006 seconds