A new facial recognition tool offers a rare look at how machines see the world. ImageNet Roulette, public tool built in an artist's studio, will take anyone's photo and apply labels to what it sees.

The tool pulls its labelling information from ImageNet, a database containing over 14 million hand-tagged images.

ImageNet Roulette is not a commercial product -- it was built in an art studio and is on display in the Fondazione Prada Osservertario museum in Milan. Artist and researcher Trevor Paglen and his team created the tool.

Paglen has created other artwork that focuses on artificial intelligence, privacy, and digital security. With ImageNet Roulette, he wanted to focus on the imperfect and sometimes discriminatory tags applied to photos in ImageNet and other databases, which are often built on top of hand-tagged data outsourced to contractors.

From the ImageNet Roulette page:

ImageNet contains a number of problematic, offensive and bizarre categories - all drawn from WordNet. Some use misogynistic or racist terminology. Hence, the results ImageNet Roulette returns will also draw upon those categories. That is by design: we want to shed light on what happens when technical systems are trained on problematic training data. AI classifications of people are rarely made visible to the people being classified. ImageNet Roulette provides a glimpse into that process – and to show the ways things can go wrong.