AI Orthopraxy: Towards a Framework for That Promotes Fairness

Abstract

This paper introduces the term AI Orthopraxy as the correct practice of AI and a framework that aims to unify some aspects associated with AI ethics. These include standards, legal, and measures of fairness. We draw from existing tools that have been peer-reviewed by academics and discussed in recent literature to provide a mechanism for assessing the level by which a model or AI technology follows the correct practices of ethical AI. This paper describes a preliminary, ongoing, study and shows the early stages of a prototype framework, including a visual representation of the level of AI Orthopraxy of a model using hive plots. This work can potentially help create fair and trustworthy AI built upon the core tenets of accountability, transparency, and fairness. One of the current limitations is that it requires validation of peers that are willing, able, and trained to evaluate an AI model or technology using standards and other novel frameworks.

Ähnliche Arbeiten