Strategic Classification in the Dark

TL;DR

This paper explores the impact of opacity in strategic classification, analyzing how hiding the classifier affects agents' behavior and the overall prediction error, and proposes transparency as a beneficial policy under certain conditions.

Contribution

It generalizes strategic classification to opaque scenarios, introduces the concept of the price of opacity, and provides conditions when transparency improves classifier robustness.

Findings

Opacity can increase prediction error in strategic settings.

Transparency can be beneficial when the price of opacity is positive.

Hardt et al.'s robust classifier is affected by opacity.

Abstract

Strategic classification studies the interaction between a classification rule and the strategic agents it governs. Under the assumption that the classifier is known, rational agents respond to it by manipulating their features. However, in many real-life scenarios of high-stake classification (e.g., credit scoring), the classifier is not revealed to the agents, which leads agents to attempt to learn the classifier and game it too. In this paper we generalize the strategic classification model to such scenarios. We define the price of opacity as the difference in prediction error between opaque and transparent strategy-robust classifiers, characterize it, and give a sufficient condition for this price to be strictly positive, in which case transparency is the recommended policy. Our experiments show how Hardt et al.'s robust classifier is affected by keeping agents in the dark.