Train machines (agents) to interact in environments through the use of rewards and punishments to incentivise specific results with Q and Deep Q Learning (DQN)
Doing things like using a generative model to classify things, building a number drawer, doing math, and attempting to break encryption with neural networks.