Anfonwch hwn fel neges destun: The possible and the impossible in multi-agent learning