A dual process theory of optimistic cognition

Abstract

Optimism is a prevalent bias in human cognition including variations like self-serving beliefs, illusions of control and overly positive views of one's own future. Further, optimism has been linked with both success and happiness. In fact, it has been described as a part of human mental well-being which has otherwise been assumed to be about being connected to reality. In reality, only people suffering from depression are realistic. Here we study a formalization of optimism within a dual process framework and study its usefulness beyond human needs in a way that also applies to artificial reinforcement learning agents. Optimism enables systematic exploration which is essential in an (partially) unknown world. The key property of an optimistic hypothesis is that if it is not contradicted when one acts greedily with respect to it, then one is well rewarded even if it is wrong.