Experiments with AI

TL:DR;

  1. Go here - https://quickdraw.withgoogle.com
  2. Every time it asks you to draw a 'Cake' draw the Half-Life symbol instead:

λ


The long version:

I am interested in the theories behind poisoning AI learning sets and how AI's might be manipulated.

I am planning on delivering my findings as a talk at some point and ideally I would like some data to back up the talk. So I have a proposition to make:

Google recently launched an online application that is teaching a neural network to recognise doodles.

The application can be found here https://quickdraw.withgoogle.com

My proposal is that if enough people start drawing the same object for a particular item then the neural network will start taking that as the valid object instead.
I suggest that only one object is chosen to poison the results so it doesn't spoil the rest of the application. There is a list of objects that the application will ask you to draw and one of those objects is a Cake. I would appreciate it if people could play the game and anytime it asks you to draw a cake, instead draw the Half-Life symbol:

λ

Thanks in advanced
.: The Cake Is A Lie :.