A simple forecasting game

摘要

We consider a population of Boolean agents playing a simple forecasting game, in which the goal of each agent is to give a correct forecast of the future state of its neighbors. The numerical results show that by using a simple inductive learning algorithm the agents are able to accurately achieve the goal of the game. However, this remarkable performance has an unexpected consequence: by learning to forecast the future, the agents dynamics freezes up at the end of the game; the only way to regain their dynamics is to forget what they have learned.