Algorithmic classification and strategic effort

Abstract

In this letter, we summarize our recent work examining the incentives produced by algorithmic decision-making. Drawing upon principal-agent models in the mechanism design literature, we construct and analyze a model of strategic behavior under algorithmic evaluation. We characterize which behaviors can be incentivized by any reasonable mechanism, showing that simple linear mechanisms are sufficient to incentivize desired behavior. However, we find that it is computationally hard to optimize even simple objectives subject to the constraint that the resulting linear mechanism induces the desired incentives.