Supplementary Material for "Interacting with fallible AI: Is distrust helpful?"
Detailed information about the stimulus material: Inspired by https://whichfaceisreal.com, the fake imagery of faces was taken from the website https://thispersondoesnotexist.com. For more information on the generation: see https://whichfaceisreal.com and Karras (2019). We took a random sample of 54 images by a scripted refreshing and saving of the images. The real images of faces were taken from the flickr-faceHQ-database (Karras, 2019). We took a random sample of 54 faces out of 70,000 available ones. As a second subset, images showing objects, landscapes, or animals were used. All of the generated images (in most cases 4) per prompt were saved. The first author made a selection based on visual inspection to remove outlier images. The real objects were taken from the THINGS database (Hebart et al., 2019). Images with a resolution below 800x800px were removed. By visual inspection further potential problematic images (unknown, hardly identifiable objects) were removed. From this set of images we chose the subset for which both a real and a fake exists.
Parts of the stimuli used for the experiments originate from whichfaceisreal.com and flickr-faceHQ-database (Karras, 2019). If you use the experiments or the material, make sure to reference this work.
Karras, T. (2019). A style-based generator architecture for generative adversarial networks. arXiv preprint arXiv:1812.04948 Hebart, M. N., Dickter, A. H., Kidder, A., Kwok, W. Y., Corriveau, A., Van Wicklin, C., et al. (2019). Things: A database of 1,854 object concepts and more than 26,000 naturalistic object images. PloS one 14, e0223792