Causal Reasoning from Meta-reinforcement Learning

TL;DR

This paper demonstrates that causal reasoning can emerge from meta-reinforcement learning, enabling agents to perform interventions, infer causality, and make counterfactual predictions in complex environments.

Contribution

It shows that model-free meta-reinforcement learning can lead to emergent causal reasoning abilities in agents, a novel approach compared to traditional formal algorithms.

Findings

Agents can perform causal inference from observational data.

Agents can select informative interventions to learn causal structures.

Agents can make counterfactual predictions in novel environments.

Abstract

Discovering and exploiting the causal structure in the environment is a crucial challenge for intelligent agents. Here we explore whether causal reasoning can emerge via meta-reinforcement learning. We train a recurrent network with model-free reinforcement learning to solve a range of problems that each contain causal structure. We find that the trained agent can perform causal reasoning in novel situations in order to obtain rewards. The agent can select informative interventions, draw causal inferences from observational data, and make counterfactual predictions. Although established formal causal reasoning algorithms also exist, in this paper we show that such reasoning can arise from model-free reinforcement learning, and suggest that causal reasoning in complex settings may benefit from the more end-to-end learning-based approaches presented here. This work also offers new strategies for structured exploration in reinforcement learning, by providing agents with the ability to perform -- and interpret -- experiments.