The re-identification game!
The concept behind CrowdOfOne is to demonstrate what re-identification is in a simple and understandable way. The game is similar to that of Guess Who?, except instead of re-identifying someone by their appearance, we are doing it via their data.
We generated a collection of fictitious people and their corresponding data attributes. The names were generated at random and the photos come from the NVidia StyleGan a generative adversarial network that creates realistic, but fake, photos of people. For further information see the paper A Style-Based Generator Architecture for Generative Adversarial Networks by Tero Karras, Samuli Laine, and Timo Aila. All of the characters and data within the game are fictitious.
For each character we generated 5 data attributes and a corresponding time, based on the types of data we know is commonly collected by organisations. The attributes represent a tiny snapshot of the type of data collected about us on a daily basis. For each of these pairings we ensured that there were at least 12 people who shared the same values. This would typically be referred to as the k-value in k-anonymity. The idea behind k-anonymity is that people are lost in a crowd. However, as the game shows, if each attribute is treated independently - as so often happens - someone might be lost in the crowd for one attribute, but in a crowd of one when looking at a set of all, or some, of their attributes together.
The game can be played by two people, currently the website is standalone, in future we hope to make it available to play online against each other. To play, two people should each open the website on their own computer and click the Play button.
Once the page loads, click "Draw Card" to get allocated a character at random. Your task is to be the first person the re-identify your opponents character. To do so you need to ask whether their character has a particular data attribute. You will see the possible attributes on the left hand side. When you click on one you will see the different time windows available for that attribute. Select one and then ask your opponent.
For example, if you selected Bought a Coffee and between 8:00 am and 10:20 am you would ask, did you buy a coffee between 08:00am and 10:20am?. Your opponent will refer to their character and answer either Yes or No. You click the relevant button on your screen and you will see the remaining people filter down to those still eligible - much like folding down the characters in Guess Who?
Once you have just one remaining person you can guess the character, first to correctly guess their opponents character wins the game!
Below are a range of articles, papers and reports that provide more information on re-identification.