Most of the discourse on algorithmic decisionmaking, whether it comes in the form of praise or warning, assumes that algorithms apply to a static world. But automated decisionmaking is a dynamic process. Algorithms attempt to estimate some difficult-to-measure quality about a subject using proxies, and the subjects in turn change their behavior in order to game the system and get a better treatment for themselves (or, in some cases, to protest the system.) These behavioral changes can then prompt the algorithm to make corrections. The moves and countermoves create a dance that has great import to the fairness and efficiency of a decision-making process. And this dance can be structured through law. Yet existing law lacks a clear policy vision or even a coherent language to foster productive debate.
This Article provides the foundation. We describe gaming and countergaming strategies using credit scoring, employment markets, criminal investigation, and corporate reputation management as key examples. We then show how the law implicitly promotes or discourages these behaviors, with mixed effects on accuracy, distributional fairness, efficiency, and autonomy.
Jane Bambauer & Tal Zarsky,
The Algorithm Game,
Notre Dame L. Rev.
Available at: https://scholarship.law.nd.edu/ndlr/vol94/iss1/1