Learning in Interactive Decision-Making: The Interplay Between Cognitive Abilities and the Strategic Environment.

Abstract

A remarkable feature of human intelligence is the ability to optimize our decisions based on the potential actions of others. This ability, i.e., strategic sophistication, is crucial in strategic interactions, where we need to predict others' actions (first-order beliefs), anticipate others' beliefs about our own possible actions (second-order beliefs), and optimize decisions based on such beliefs. While behavioral research has highlighted systematic departures from theoretically optimal behavior in strategic interactions, little is known about the possibility of enhancing strategic sophistication. In particular, no studies investigated whether and how the interaction between exogenous factors (i.e., the learning environment) and endogenous factors (i.e., individual cognitive abilities) shapes learning in strategic settings. In a novel mouse-tracking study, we manipulate the learning environment and test its interaction with individual cognitive abilities in determining context-specific and transfer of learning in interactive games. Choice and process data reveal that the interplay between individual cognitive abilities and the learning environment does modulate participants' learning. The learning environment determines what is learned and whether acquired knowledge is applied in similar contexts and transferred to novel settings. Moreover, learning success in different strategic environments depends on individual cognitive abilities. In particular, higher levels of cognitive reflection are necessary to learn sophisticated strategic behavior (i.e., forming second-order beliefs) and transfer acquired knowledge to more complex strategic environments after receiving relevant feedback. However, higher cognitive reflection levels are insufficient to prevent the misapplication of procedures learned in a specific environment to other strategic contexts with substantial structural differences. Our results provide novel insights into the factors that promote or hamper learning in interactive decision-making.