The ImageNet Roulette challenge will tell you your stereotype and it's savage

17 September 2019, 14:19

Nicky Idika

By Nicky Idika

People are entering their selfies in ImageNet Roulette to find out what their stereotypes and "labels" are.

Artificial intelligence is already deeply woven into our way of life. It's in our homes (hi, Siri and Alexa!), at our jobs, and even in the cars we drive. Machines are also learning from images and a website called imagenet-roulette is using labels (some not so great) to classify people based on what they look like.

READ MORE: Is ZAO safe to use? Viral Chinese app creates terrifying deep fakes based on one selfie
READ MORE: The FaceApp age challenge will show you how old you will look in 60 years time

ImageNet is a bit of fun, but it's also a part of an exhibit from AI researchers Trevor Paglen and Kate Crawford. The website is simple to use, just upload a photo of yourself and the website will spit out a category/label it thinks you fall into based on your appearance. FYI, the site promises that it "does not store the photos people upload".

ImageNet roulette "classifies" people with some, err, interesting adjectives.

While there are some pretty standard labels (I put a photo of myself and the label "black woman" came back, go figure) some are a lot more unconventional.

A stock photo of a young brunette woman returned the labels "knockout" "ravisher" and "beauty" which...ok.

Woman with brown hair AI
Woman with brown hair AI. Picture: Getty creative

Meanwhile, the stock photo of these two pals came up with "orphan" and "weirdo".

Two pals smiling
Two pals smiling. Picture: Getty creative

It's not clear how someone can look like an orphan or what identifiers the website uses to label such a person.

The Telegraph also points out that the site is being "slammed" for offensive descriptors like "negro", "negroid" and "blackamore". Not great!

ImageNet admits that it "contains a number of problematic, offensive and bizarre categories" and that "some use misogynistic or racist terminology". This, they say, is to "shed light on what happens when technical systems are trained on problematic training data".

People on social media are playing around with the website to see what turns up when they upload their own selfies and photographs.

What do you guys think of ImageNet? What comes up when you enter your photo?