Learn Like A Machine is an interactive web game where you learn how machine learning is operated by a computer. In the game, you play as an avatar character that represents the computer itself. The mission of the avatar is to learn as much as it can before moving on to the next stage of its learning journey. Each fruit stands for pieces of information that comprise the whole. For example, once the avatar has “eaten” all the apples, it has analyzed and learned that what it has been eating are apples. What also comes with each fruit at each stage are background graphics. It is exactly as it is, they are background representations of environments that the information lives in. In this world, information is mixed in as noises with other fragments of other information. A good example is an apple on a countertop. People are good at recognizing that the apple is physically sitting on another object which is the countertop. Given this example, for the information to internalize successfully within the computer, the machine needs to recognize different apples among different background graphics. When the avatar has reached the final fruit and stage, the cycle of learning loops and the computer continues to learn. This cycle is modeled off of how people are always learning.
I was inspired by Forensic Architecture Triple-chaser project where they utilized 3D modeling to place tear gas grenades in generated backgrounds and then used machine learning to have the algorithm identify those grenades.
Another project Slight Street Sign Modifications Can Completely Fool Machine Learning Algorithms” is a really interesting read. Small graphic changes to a street sign can fool the machine learning algorithms into reading the sign as saying something different.