AI Governance mapped

Doug Stauber
IBM Data Science in Practice
3 min readMay 21, 2024

watsonx.governance has been touted as the most comprehensive AI Governance tool as it covers multiple types of AI Governance requirements all in a single platform.

But what does that mean?

AI Governance is an overloaded term, and different organizations claim to support AI Governance “end to end” while actually only covering a subset of required functionality. IBM covers three different types of AI Governance all within a single platform.

Mapping the types of AI Governance found within watsonx.governance

Let’s break down these three types:

1. Model Risk Governance & Regulatory Compliance

This portion of AI Governance is the most human-centric. It organizes personas, from risk validators to legal and compliance officers, into an organized AI team. It provides dashboard views, so all stakeholders of AI can see the latest on AI efforts including high level views of all models in production down to individual metrics for a single model. It also provides process governance: workflows with approvals to ensure the steps supporting AI are covered. This includes the ability to bring in a human at the right time without babysitting of models. For example, if a model begins to drift while in production, the system will notify a model validator to investigate; if an issue is found, the system will then notify the appropriate data scientist for further investigation. In this way, the system can enforce both internal policies and external regulations for consistent execution for all AI efforts enterprise wide. Finally, all efforts are logged and tracked for audit and reporting purposes. In short, Model Risk Governance & Regulatory Compliance creates an organized AI team. They are armed with the right dashboards and workflows to get the latest status of AI while enforcing polices and regulations with an audit trail.

2.AI Artifact Governance

This portion of watsonx.governance is about documentation. AI efforts, whether generative AI or predictive ML, are organized into use cases. These use cases are tracked and mapped to stages… from development to production. This creates a central inventory to document AI and ML efforts, including the ability to assign an overall risk score for each use case. This model registry is enhanced by an AI Factsheet. AI Factsheets are always up-to-date model cards. These Factsheets include information such as data scientist owner, model built, test data, and latest metrics. And unlike other model cards, they are updated automatically via a one-time notebook configuration. Put simply, AI Artifact Governance documents how AI models map to specific use cases while providing always up-to-date Factsheets with deep detail about each AI model.

3. Metric Computation for Development and Validation Governance

watsonx.governance is a AI calculation engine. It evaluates Predictive ML and Generative AI both at development time and continuously in production for a variety of metrics. It can calculate quality scores including accuracy, drift, and model health metrics. It can calculate safety scores for toxic language and personal identifiable information of LLMs. And, it can calculate responsible AI metrics for explainability, attribution, and fairness. The metadata watsonx.governance calculates helps to inform data scientists, prompt engineers, and other stakeholders on the status of AI. This metadata is then passed into the other stages of AI Governance providing a single shared source of truth.

1+2+3 = Comprehensive

Together, these three types of governance cover the full end-to-end of the AI Lifecycle from use case definition to model retirement. Plus they span a large variety of stakeholders, from more business users of process governance to the data scientists using metric computation. In this way, watsonx.governance offers the largest breadth of capabilities covering the full range of AI lifecycle and stakeholders.

In addition, we are constantly expanding our scope of AI Governance. Watch for more exciting announcements on how we are extending our capabilities in the coming months.

Learn More and Try Today

To learn more about watsonx.governance or give it a try for free, visit our website at https://www.ibm.com/products/watsonx-governance

Here you can start a trial or setup a briefing where you can request a demo of the above features in action.

#IBM #watsonx @AIGovernance

--

--