The Double-Edged Sword of Behavioral Responses in Strategic Classification: Theory and User Studies

  • 2024-10-23 18:42:54
  • Raman Ebrahimi, Kristen Vaccaro, Parinaz Naghizadeh
  • 0

Abstract

When humans are subject to an algorithmic decision system, they canstrategically adjust their behavior accordingly (``game'' the system). While agrowing line of literature on strategic classification has used game-theoreticmodeling to understand and mitigate such gaming, these existing works considerstandard models of fully rational agents. In this paper, we propose a strategicclassification model that considers behavioral biases in human responses toalgorithms. We show how misperceptions of a classifier (specifically, of itsfeature weights) can lead to different types of discrepancies between biasedand rational agents' responses, and identify when behavioral agents over- orunder-invest in different features. We also show that strategic agents withbehavioral biases can benefit or (perhaps, unexpectedly) harm the firm comparedto fully rational strategic agents. We complement our analytical results withuser studies, which support our hypothesis of behavioral biases in humanresponses to the algorithm. Together, our findings highlight the need toaccount for human (cognitive) biases when designing AI systems, and providingexplanations of them, to strategic human in the loop.

 

Quick Read (beta)

loading the full paper ...