In my most challenging reinforcement learning project to date, we throw DQNs, PPOs, and autoencoders at Tetris to see whether AI can learn the kind of spatial reasoning that comes so naturally to humans.

Instead of playing pictionary with your friends, now you can play it with AI. And after you've played it, you can read about how it was made.

A foray into an early form of reinforcement learning, the Deep Q Network. After much teeth-pulling, tweaking, and training, our hero eventually learns out how to play a simple game with mediocre competence.

Simulating the evolution of neural networks through natural selection by pitting thousands of predators and prey against one another -- and finally proving Charles right.