CIF: Small: Learning in Strategic Environments with Applications in Algorithmic Fairness

With the proliferation of predictive models in consequential decision-making roles, their collective impact on society can no longer be ignored. In this project, the investigators aim to develop statistical tools to assess the societal impacts of predictive models and policy recommendations in order to improve social welfare. Technical activities focus on a special class of prediction problems in which the predictive model itself affects the problem environment - this is a phenomenon known as performativity, and such performativity is often caused by the strategic behavior of agents in the environment.

For example, spammers (the agents) will change the content of their messages to evade spam filters (the prediction models). The investigators aim to leverage performativity to improve social welfare and reduce objectionable disparities among the agents. Broader-impact aspects of the project include application of the technical work to steering labor markets and social-media platforms with possible extension to the criminal-justice, sustainability, and transportation sectors.

Towards this goal, the project intends to develop methods that enable artificial-intelligence (AI) practitioners to learn how their predictive models affect the problem environment in order to overcome barriers to the leveraging of the impacts of predictive modeling for social good. Armed with knowledge of how predictive models affect the environment, the investigators plan to develop theories and methods to help practitioners leverage this knowledge to steer the environment to improve social welfare.

Rather than treating performativity as a problem to be overcome as is usually the case, the investigators instead leverage performativity to circumvent long-standing barriers to the broader adoption of algorithmic-fairness interventions, aligning the incentives of (possibly non-altruistic) organizations and users in sociotechnical systems so as to escape from impossibility-of-fairness results.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.