What Is ImageNet Roulette?

ImageNet Roulette is a web-based app that allows users to upload an image of themselves and then see what results come up when the app runs the image through its machine learning algorithm. The app was created by artist Trevor Paglen and programmer-artist Dan Shiffman.

The goal of the app is to show how machine learning algorithms can be biased, and to start a conversation about why that bias exists.

When you first visit ImageNet Roulette, you’re presented with a blank Canvas and the instruction to “Upload an image.” Once you select and upload an image, the app runs it through its machine learning algorithm.

The results are displayed on the screen, along with the percentage confidence that the algorithm has in its results.

The app pulls its data from ImageNet, a database of images that’s often used to train machine learning algorithms. ImageNet was created by Stanford University in 2009 and contains more than 14 million images.

It’s organized into more than 20,000 different categories, each of which has thousands of images.

The app doesn’t just display the results of its machine learning algorithm; it also shows how those results compare to what humans would classify the same image as. For example, if the app thinks an image is of a “monkey,” it will also show how confident humans are that the image is of a monkey.

In many cases, the app’s results are very close to what humans would classify. But in some cases, there are significant differences.

The goal of ImageNet Roulette is to start a conversation about why these biases exist in machine learning algorithms. The developers hope that by showing how these algorithms can be biased, we can start to understand why that bias exists and what we can do about it.