Lying for the Greater Good: Bounded Rationality in a Team

The article is concerned with the interaction between fully and boundedly rational agents in situations where their interests are perfectly aligned. The cognitive limitations of the boundedly rational agent do not allow him to fully understand the market conditions and lead him to take non-optimal decisions in some situations. Using categorization to model bounded rationality, we show that the fully rational agent can manipulate information to help decreasing the expected loss caused by the boundedly rational agent. Assuming different types for the boundedly rational agent, who differ only in the categories used, we show that the fully rational agent may learn the type of the boundedly rational agent along their interaction. Using this additional information, the outcome can be improved and the amount of manipulated information can be decreased. Furthermore, as the length of the interaction gets longer the probability that the fully rational agent learns the type of the boundedly rational agent increases.