Are GNN+RL methods superior to existing algorithms for economic computational problems?

Establish whether graph neural network and reinforcement learning methods are necessarily superior to existing methods for solving a diversity of computational problems in economics.

Background

The paper introduces a demonstration of graph neural networks (GNNs) combined with reinforcement learning (RL) for a toy infrastructure-upgrade problem and compares them to other approximate methods. While the demo highlights scalability advantages, the authors explicitly refrain from claiming superiority over established approaches.

They note that determining superiority across economic computational problems is an early-stage, active research topic and explicitly state that it remains to be shown whether these methods are necessarily superior. This frames a broad open question about performance dominance and applicability across tasks in economics.