# Superstition in the Network: Deep Reinforcement Learning Plays Deceptive Games

## TL;DR

This paper investigates the failure modes of deep reinforcement learning agents, specifically A2C, on deceptive video games, revealing their vulnerabilities and contrasting them with planning agents to understand their limitations.

## Contribution

The study introduces a set of deceptive games to test deep RL, compares their performance with planning agents, and proposes a typology of deceptions to analyze RL failures.

## Key findings

- Deep RL agents are reliably deceived by certain games.
- Failures of RL differ from planning agents, highlighting unique shortcomings.
- A typology of deception modes in RL is proposed.

## Abstract

Deep reinforcement learning has learned to play many games well, but failed on others. To better characterize the modes and reasons of failure of deep reinforcement learners, we test the widely used Asynchronous Actor-Critic (A2C) algorithm on four deceptive games, which are specially designed to provide challenges to game-playing agents. These games are implemented in the General Video Game AI framework, which allows us to compare the behavior of reinforcement learning-based agents with planning agents based on tree search. We find that several of these games reliably deceive deep reinforcement learners, and that the resulting behavior highlights the shortcomings of the learning algorithm. The particular ways in which agents fail differ from how planning-based agents fail, further illuminating the character of these algorithms. We propose an initial typology of deceptions which could help us better understand pitfalls and failure modes of (deep) reinforcement learning.