AI Governance Terminology Guide

AI Governance

The management and monitoring of responsible AI technology use to understand where AI is used, built, or procured across the business for risk mitigation, legal and regulatory compliance, and stakeholder trust.

AI System

An object type used to assist organizations in capturing essential details on regulatory-mandated information, promoting, compliance, and facilitating easier audits. This differs from traditional IT assets, and understanding that different systems vary in their levels of autonomy and adaptiveness after deployment, this inventory can track version changes, model updates, and performance shifts over time.

Dataset

An object type to keep track of a specific dataset used for building, training, or testing a model. It simplifies how we understand and manage the data used by AI systems. With this, organizations can understand how the data used by AI is tracked, versioned, and annotated in the data source.

Demographic Parity Difference

A fairness metric to assess how much disparity exists between models from different groups, so each group is treated fairly.

Demographic Parity Ratio

Ratio of selection rates between different groups in a machine-learning model.

Equalized Odds Difference

A fairness metric to assess if the machine-learning model performs equally across different groups.

Equalized Odds Ratio

Ratio of error rates across groups for predictions.

Generative AI

Often used in art, text, and design, this AI technology leverages machine-learning models to learn patterns and relationships to generate new content.

Inventory Record

A detailed record of the given object type that contains basic details, related assessments, risks, controls, documents, links, and tasks.

Model

The learning framework used to build and inform the AI/ML application. This object type helps users distinguish whether the model is of internal origin or from a third party, what risks stem from the use of the systems, and for what tasks the model is intended, so organizations can properly identify how many models are in production.

Project

An object type used to track where AI is deployed for business initiatives, applications, or services to achieve specific business outcomes. This inventory allows organizations to capture a detailed record of a system’s primary optimization goal(s), purposes, or foreseeable uses of the AI system to improve project tracking and progress monitoring.

Relationship Record

A detailed record to better understand data flows between associated inventory objects.