These ‘artifacts’ are certainly very strange semantically (if you consider everything on the internet is what the label says it is). While the juxtaposition is

These ‘artifacts’ are certainly very strange semantically (if you consider everything on the internet is what the label says it is). While the juxtaposition is referentially alarming, and has a pattern, that does not prove the nature of the intent as racist. In other words the seemingly easy pattern for the auditor/audience could also be an easy formula for the maker—especially if it were a bot. Does anyone know the state of the art for auto-editing and related auto-labeling to say that such content would be easy for a bot to assemble. An easy—excuse the expression— ‘toy’ example? Not knowing much about the domain, but I also imagine this content could also be manually edited experiment for labeling experiments? Relatedly, I recall early Twitterbot experiments. They were usually silly, strange, and all had a clearly detectable conceit. I’m still subscribed to one called Archillect (https://twitter.com/archillect https://twitter.com/archillect). Early days I recall most images were very dark—both monochromatic and thematically. Their "art" would be the subjective evaluation of a visual with it's interpretation ('being' and 'seeming' dark both visually and thematically--lots of anime or high-tech, and not much 'nature'). I’m not saying this isn’t the work of a @$$h@l3. EDIT: edited my description of Archillect as artistic experiment.