Colour versus Shape Goal Misgeneralization in Reinforcement Learning: A Case Study

NeurIPS 2023 Workshop ATTRIB Submission16 Authors

Published: 27 Oct 2023, Last Modified: 08 Dec 2023ATTRIB PosterEveryoneRevisionsBibTeX
Keywords: deep reinforcement learning, AI safety, goal misgeneralization, behaviour attribution, behavioural outliers
TL;DR: Colour versus shape goal misgeneralization in Procgen Maze happens mainly through colour alone, and this preference can change by only changing the training random seed.
Abstract: We explore colour versus shape goal misgeneralization originally demonstrated by Di Langosco et al. (2022) in the Procgen Maze environment, where, given an ambiguous choice, the agents seem to prefer generalization based on colour rather than shape. After training over 1,000 agents in a simplified version of the environment and evaluating them on over 10 million episodes, we conclude that the behaviour can be attributed to the agents learning to detect the goal object through a specific colour channel. This choice is arbitrary. Additionally, we show how, due to underspecification, the preferences can change when retraining the agents using exactly the same procedure except for using a different random seed for the training run. Finally, we demonstrate the existence of outliers in out-of-distribution behaviour based on training random seed alone.
Submission Number: 16