Ambiguous Strategic Classification

A common assumption in strategic classification is that the classifier is public knowledge. However, it remains unclear whether, and why, a system would choose to commit to full disclosure. We study a setting in which regulation requires the system to disclose some, but not all, of the information. This induces a learning task in which the learner must jointly optimize the classifier and the uncertainty surrounding it. To this end, we adopt from robust mechanism design the notion of ambiguity, which in our setting allows the learner to reveal a set or range of possible classifiers, while privately choosing which of them to ultimately realize. We investigate how ambiguity affects the learning task, develop efficient algorithms for computing best-responses and training, and empirically explore strategic learning and its outcomes in this novel setting and using our approach.

Strategic classification studies learning in setting where individuals can adapt their behavior in response to the classifier in order to obtain more favorable outcomes. Research in this area commonly assumes that the classifier is fully known to the users affected by it. However, it is unclear whether real systems would choose to reveal all of this information, or whether regulation would even permit it. In this work, we study a setting in which the system is required to disclose only partial information about how decisions are made. Rather than committing publicly to a single classifier, the system can reveal a set or range of possible classifiers while privately choosing which one to apply. This creates ambiguity for individuals attempting to strategically adapt their behavior. We introduce a framework for studying learning in the ambiguous strategic setting. We develop efficient algorithms for training and strategic response computation, analyze their performance, and experimentally investigate how ambiguity affects strategic behavior as well as learning and social outcomes.