‘Nerd,’ ‘Nonsmoker,’ ‘Wrongdoer’: How Might A.I. Label You?
This article discusses the digital art project ImageNet Roulette, a museum exhibit with a (now defunct) companion website that examines pictures of faces that participants submit to it and then applies tags to them based on the data it was trained with, which is a compendium of tagged images called ImageNet that was originally created by Stanford University. Some of the tags in the ImageNet database are innocuous (they include things like “cheerleaders” or “welders”) but some are judgmental, including “failure” and “slut.” Even more troubling is the fact that the way the pictures are tagged often reflects racist, sexist, or homophobic stereotypes. Since the ImageNet database is widely used by tech companies for training artificial intelligence programs, the creators of the art installation wanted to attract attention to the way that we can imprint bias into artificial intelligence systems even without our knowledge.
This article relates to the concepts we’ve discussed in class due to the element of racial bias that the AI program shows—as an example, one African-American man who submitted his photo to the website received the tags “wrongdoer” and “offender.”
This event is very important because it shows how biases of several varieties are included in the ImageNet database, which is widely used to train AI, in addition to shedding light on how AI can reflect societal structures of power even when this is not intended. This matters because, since AI is becoming more widely adopted by companies, police departments, and governments, bias present in AI can lead to yet more unfair treatment of those who already lack privilege. Hopefully, this exhibit will raise people’s awareness of this issue and encourage companies and universities to create less biased data sets to train AI.
Personally, I appreciate this project, since it is working to help people become aware of the fact that new technologies are not always going to be immune to old prejudices; we need to be vigilant and make sure that we always work to fight bias. Regarding the educational applications of this article, though I would not use it myself since it’s not in Arabic, I think it could be used in an Advanced Low level English class (I think it needs to be Advanced due to how abstract the article’s topic is) as a starting point for a discussion on the social justice implications of modern technology.
I am very grateful for the knowledge and concepts you have generously shared with me. We appreciate your strength.
Geometry Lite is a game that can be enjoyed by players of all ages and skill levels.
This article is a valuable contribution to the field and a pleasure to read.