I came across this work on MARL by Google Deepmind and wanted to share it with the class [1]. They showed how agents in their environment could simulate microeconomic behaviours, namely production, consumption, and trading. This environment, called Fruit Market, was a place where agents could engage in trading. Over time, the agents learned to make rational choices regarding the economy and the supply and demand changes that were taking place. Some agents also learned how to arbitrage by transporting goods based on the potential profits. The full paper is available, but it is pretty long and detailed. I have linked it in case anyone wants to take a look [2].

The Deepmind team said that this work is a step towards achieving AI by taking in aspects of social intelligence. 

Sources:

[1] “Emergent Bartering Behaviour in Multi-Agent Reinforcement Learning,” Google DeepMind. Accessed: Dec. 23, 2023. [Online]. Available: https://deepmind.google/discover/blog/emergent-bartering-behaviour-in-multi-agent-reinforcement-learning/

[2] M. B. Johanson, E. Hughes, F. Timbers, and J. Z. Leibo, “Emergent Bartering Behaviour in Multi-Agent Reinforcement Learning,” 2022, doi: 10.48550/ARXIV.2205.06760.