In this case, a simple random black rectangle on top of blank canvas.
You're shown two random ones.
Keep voting for the one you think that you like more, and you'll be training a neural net to help decide for a better one over time.
Often, we might not know which are all the rules that make us like a piece more than another, but with this method, you don't need to know...just have to keep voting and it'll be used as training data to train a simple neural net (using brain.js Javascript Library) on the page.
and over time it gets better at acting as you (or like you would vote) and each time you click generate() it can act as you and vote like 1000 times (after giving enough real votes of course).
program page: https://thelotteryforum.com/braint.html
Youtube demo after my 300 votes or so.

It's in the discovery and testing phase.
In the future I want to change the code to change the input of training data to take in more variables which would represent more complex than a simple black rectangle.

Sorry about earlier posted wrong video so It was unavailable for a while

This is a voting or election method to train a neural net. It's probably fairly predictable what humans will vote for. My prediction is it will produce something like the Mona Lisa (once the blockier pixels have been voted-out )

Why not just use existing elections to decide? Allocate a random decision to each candidate and then see who won. IMHO humans are extremely good a electing complete idiots. Here they are only electing a piece of art and can't self-harm like they usually do

Oh no not Mona Lisa, this is just to get the most fitting black on white canvas.
So for this one the end will always be a black rectangle on white canvas.

As for the training data I am only feeding it top left corner coordinate, width and height of black rectangle. (Probably shouldn't have called it pieces of art, but just black rectangles on white which could be art hehe)

In future pages I plan to feed it more data for other shapes or combinations of shapes

This is what the trained neural net thinks I would vote for (with 249 votes as training data).
I generated 1000 times for each tile and did it 9 times/tiles and composed this one.

comments are interesting on this one: https://old.reddit.com/r/generative/com ... ectangles/

This one is 9 pairs of 1 black rectangle and another with inner white rectangle.
Trained with 300 votes.