Based on 2048 by Gabriele Cirulli – https://github.com/gabrielecirulli/2048
This AI uses a Deep Learning Artificial Neural Network with reinforcement learning for training to teach itself how to play 2048. At the outset it only knows that there are four possible choices, to swipe up, down, left and right. It learns that it likes to get high scores per swipe. It also likes to have empty squares on the board, and it likes to have high numbers on the board. It also enjoys having a long game time.
I have not currently been able to achieve a 2048 with this AI, but that will come with a proper neural network design, which I am constantly updating.