jonathanbenz/classifier

Short game that brings awareness to the subject of racist, sexist, and homophobic/ transphobic classifications used in facial recognition software.

Classifier

This was the final project I created for my art history class which focused on topics such as identity, portraiture, and facial recognition software.

The gist of this interactive video game is to highlight and bring awareness to the fact that many popular facial recognition datasets have contained classifications which are extremely biased and prejudiced against minority groups. Datasets such as Imagenet, 80 Million Tiny Images, and Tencent-ML-Images have all been exposed to contain offensive classifications which are racist, sexist, and homophobic/ transphobic. I bet there are many more datasets that contain offensive terms in them too, probably unknowingly, because many modern datasets were unfortunately built off these offensive datasets and grandfathered these terms into their own.

In this game, the user plays as an AI being trained to classify aliens by an evil researcher. At first, the classifications that the researcher uses on the aliens is pretty objective and not problematic. But as you progress through the narrative, it is clear that the researcher has his own agenda and is trying to implement his own biases and prejudices against specific types of aliens. As the user gets trained further and further, you realize how morally wrong the researcher is.

Even though the game is quite light-hearted, it is important to note that the classifications that the researcher feeds to the AI are based off of real life classifications found in AI facial recognition datasets...

Huge shoutout to Veeya, the artist. Her aliens are quite lovely!