How to Train your Decision-Making AIs
How do humans transfer their knowledge and skills to artificial decision-making agents more efficiently? What kind of knowledge and skills should humans provide and in what format?
The combination of deep learning and decision learning has led to several impressive stories in decision-making AI research, including AIs that can play a variety of games (Atari video games, board games, complex real-time strategy game Starcraft II), control robots (in simulation and in the real world), and even fly a weather balloon. These are examples of sequential decision tasks, in which the AI agent needs to make a sequence of decisions to achieve its goal.
Today, the two main approaches for training such agents are reinforcement learning (RL) and imitation learning (IL). In reinforcement learning, humans provide rewards for completing discrete tasks, with the rewards typically being delayed and sparse. But, success stories about RL and IL are often based on the fact that we can train AIs in simulated environments with a large amount of training data.
What if we don’t have a simulator for the learning agent to fool around in? What if these agents need to learn quickly and safely? What if the agents need to adapt to individual human needs? These concerns lead to the key questions we ask, which are: How do humans transfer their knowledge and skills to artificial decision-making agents more efficiently? What kind of knowledge and skills should humans provide and in what format?