This machine tracks visitors to its exhibit, recording their race and sex. When a unique face enters its range of view, it records this data. It makes a tally on a piece of paper divided into six regions comprising white male, black male, asian male, white female, black female and asian female. Once the paper is full, it halts.

Intentionally or otherwise, algorithms that we use on a daily basis have been encoded carrying the biases of their creators. By 2020, technology has advanced to the point that we can programmatically detect sex, race, age, and, by combining factors, in all likelihood, class.

To what end is this technology being used?

This machine does nothing more than recording what it detects, but it very well could. We've left that part out and leave it up to the viewer to imagine what actions a machine like this could take.

Run pip install -r requirements.txt

Download the lfwa database and unzip it into lfwa

Download the color feret database and unzip it into colorferet

Run ruby scripts/feret_mutator.rb

Run scripts/extract-features

Run scripts/train

Run scripts/run

Based in part on https://github.com/wondonghyeon/face-classification