Data Agency Theory: A Precise Theory of Justice for AI Applications

Data collection methods for AI applications have been heavily scrutinized by researchers, policymakers, and the general public. In this paper, we propose data agency theory (DAT), a precise theory of justice to evaluate and improve current consent procedures used in AI applications. We argue that data agency is systematically defined by consent policies. Therefore, data agency is a matter of justice. DAT claims data agency ought to be afforded in a way that minimizes the oppression of data contributors by data collectors. We then apply DAT to two salient consent procedures in AI applications: Reddit's Terms of Service agreement and the United States's IRB protocols. Through these cases, we demonstrate how our theory helps evaluate justice and generate ideas for improvement. Finally, we discuss the implications of using justice as an evaluation metric, comparing consent procedures, and adopting DAT in future research.