Self-Destructive RL Agents

Innocent changes to RL reward functions produce surprising behaviors

Image by Stefan Keller from Pixabay.

Problem Setting

Reinforcement Learning is a popular branch of AI/ML that tries to learn optimal behavior by interacting with the world and maximizing a reward signal. For example, when playing games the agent will make a decision on how to interact with the world, and the environment game will reward accordingly.

Atari Breakout. Here the agent’s reward signal is the score of the game. Environment …

agents deep learning optimization reinforcement learning rl