Located at the Fondazione Prada museum in Milan as well as online ( https://imagenet-roulette.paglen.com ), the ImageNet Roulette art installation is part of the Training Humans exhibition. The exhibition covers the data-sets, photos and training of Artificial Intelligence.
Earlier today I was sent a link to this by a friend who requested that I submit a selfie of myself to their website. The software will then proceed to analyze the photo of my face and apply a number of labels to me.
For modesty’s sake I will not share my results here as I’m confident they are a bit more flattering than many have been receiving. You can see the photo that I submitted of myself on this page and try it yourself even.
This installation echoes the recent tide of dissatisfaction with AI’s facial recognition and applications. It is meant as a provocation to, as far as I see, demonstrate how terrible AI is at labeling people.
Some have been called “offender”, “rape suspect”, “pipe smoker” as well as far worse. The AI was trained using Mechanical Turk, paying workers fractions of a cent for each one they do. Most of these workers are located in India, East Asia and Anglo-Africa.
While I suspect the flattering labels applied to my photo were the result of me softly staring into the camera, smiling and being complimented by some lighting, people do not do this normally. Meaning as you walk about in your daily life, a regular image scan of your face will likely be labeled quite unflatteringly and of course quite inaccurately or even offensively.
I think it is a wonderful project and encourage you to try it out now.
Leave a Reply