It is being better if you using https://turbowarp.github.io/#416634264/ At beginning, just watch. The circles on left are split into groups (look at connections). The circles on right will “learn” how to telling you if they circles on left are same color or not. (White for same, black for different: brightness is probability of being same) After error is sufficiently low, another group of circles appears, and you can changing the color of the bottom left ones by click them. I would have using visual that is looking like a neural network, but there was not being enough space on screen. Sorry if directions are being difficult to understand :(
When the project is starting, the AI is not knowing what it means if the circles are same color. It “learns” what it is meaning with training data, and can giving correct answer with different input. Inspiring by @GCGamesAndCartoons
