How does an AI see the world?
In this series I’m exploring how AI algorithms could see the world. I’ve developed an algorithm that focuses on parts of the world that have a certain level of detail. By subdividing the world in ever smaller parts where information density is high enough, the algorithm creates a specific abstraction of reality. By iteratively modifiying the code, I’m trying to find a balance between the deterministic output of the algorithm and aesthetical implications thereof.
Shannon Entropy
The algorithm takes an image and divides it up in ever smaller parts. Parts that contain a lot of detail are divided more, creating a more fateful representation of that specific part of the image. In a sense, the algorithm pays attention to more detail where it matters most.
A way of measuring detail or information density is by using Shannon Entropy. This is a mathematical function that describes the amount of information or “surprise” in a system. By using this function, the algorithm can decide which parts to subdivide and which parts to leave as is.