Quantifying Curiosity: A Formal Approach to Dissociating Causes of Curiosity

Curiosity motivates exploration and is beneficial for learning, but curiosity is not always experienced when facing the unknown. In the present research, we address this selectivity: what causes curiosity to be experienced under some circumstances but not others? Using a Bayesian reinforcement learning model, we disentangle four possible influences on curiosity that have typically been confounded in previous research: surprise, local uncertainty/expected information gain, global uncertainty, and global expected information gain. In two experiments, we find that backward-looking influences (concerning beliefs based on prior experience) and forward-looking influences (concerning expectations about future learning) independently predict reported curiosity, and that forward-looking influences explain the most variance. These findings begin to disentangle the complex environmental features that drive curiosity.