AI governance is an all-encompassing strategy that establishes oversight, ensures compliance and develops consistent operations and infrastructure within an organization. It also fosters a human-centric culture.  

This strategy includes specific governance domains such as data governance and model governance, necessary for a unified AI approach.

Why AI governance matters 

The governance of AI systems has become a critical area of focus. To many practitioners, AI governance seems like a redundant concept when we already invest in data and model governance. However, the concepts are distinct and necessary for a strategic organization.

AI governance: The restaurant analogy 

To better understand AI governance, let’s compare it to running a restaurant:

• Data governance: Inventory management
• Model governance: Creating a menu
• AI governance: Overall restaurant management

Data governance: The foundation of AI systems

No restaurant is successful without inventory—produce, meats, dairy, linens, napkins, cutlery, and good customer service are all essential components of a fully functioning establishment. Most self-respecting restaurants must source their produce and meats, manage their storage and ensure they are used by the expiration date.

The restaurant must ensure it has all the necessary components to serve the meals on the menu, provide a delicious and fun experience and adhere to the health code. Inventory management is the data governance of the restaurant world.

Data governance involves managing data-related compliance needs such as tracking data lineage and metadata, enforcing data security policies, remediating data issues, applying business rules to data fields and maintaining overall data quality. Data governance is the strategic framework for managing data within an organization, focusing on policies, processes and boundaries to manage the data lifecycle effectively, just like a restaurateur would manage their inventory. Data governance ensures regulatory compliance and fosters a shared understanding of data, including lineage tracking and transparency.

This comprehensive approach to data management is foundational for any organization aiming to use data responsibly and efficiently.

Date governance focuses on questions such as: 

1. Data compliance needs: What are the requirements for data storage and usage? (What does the health code require we store our chicken?)
2. Data definitions: What specific terms and definitions are used within the organization? (When you say flour, are we talking about all-purpose flour, bread flour, or something else altogether?)
3. Data lineage: Where does the data come from? (Where did we source the ingredients from?)
4. Metadata: What are the specifics of the data? (Is this milk 2% or fat-free? Are these macintosh apples or fujis?)
5. Data issues: Are there any issues, inconsistencies or inaccuracies in the data that need to be remediated? (Is the lettuce looking wilted?)
6. Business rules: What business rules need to be applied to our data fields? (Is our restaurant vegan and is our butter also vegan?)
7. Data quality: What are the rules in place to ensure data quality? (How are we making sure our food didn’t expire in the fridge?)

Model governance: Ensuring confidence and compliance 

Model governance is like creating a restaurant menu. It involves developing and overseeing models to ensure they serve their purpose effectively, just as a menu is curated to offer a variety of dishes that align with the restaurant's theme and available ingredients.

Model governance guides the entire lifecycle of a model, including development, testing, auditing, deployment and monitoring. It aims to identify bias, enhance understandability and interpretability, protect user data and privacy and maintain model performance.

Model governance focuses on questions such as: 

1. Bias prevention: How do we ensure our models are not exhibiting unfair or unexpected bias? (How do we make sure we do not cross-contaminate our food in the kitchen?)
2. Interpretability: How do we make sure our models are understandable or interpretable? (Do our customers and staff know what a Julienne is? Do they need to know to make an educated selection about food?)
3. Monitoring: What monitoring do we need to establish to ensure the model continues to perform as expected? (What process do we need to establish in the kitchen to make sure the dish is consistent week to week?)
4. Compliance: Does the model comply with all existing and relevant laws? (Are we aligned with food safety laws?)

AI governance: Reflecting organizational values 

Having a menu and ingredients is only a part of running a restaurant. You also must manage the location and seating, marketing, hiring staff and talent management, from training new team members to ensuring your chef doesn’t quit at the peak of dinner service. AI governance provides a similar framework for organizations using data.

AI governance provides a holistic view of AI usage, anticipates potential negative impacts, and reflects organizational values. It prepares for AI use cases, assigns roles and responsibilities, and identifies tools that support AI use.

AI governance focuses on questions like:  

1. Comprehensive AI usage: Does the organization have a complete view of where AI is used in the organization? (What are all the pieces that make this restaurant operate?)
2. Risk management: How do we identify and manage risks? (How are we keeping track of food recalls, food-borne illnesses, staffing issues, etc.?)
3. Negative impacts: What negative impacts of AI solutions/use cases can be foreseen and expected?
4. Organizational values: How do our values impact our approach to AI governance? (Does our restaurant specialize in fast food or fine dining?)
5. Future AI use cases: What are the current or planned AI use cases? (What produce will be seasonal soon? Are we planning to open a drive-through or take-out counter?)
6. Governance and accountability: What group will be responsible for adapting and enforcing AI governance and accountability policies? (Who is acting as the food safety inspector?)
7. Supporting tools: What tools can be used to enable or support the use cases, values, and metrics of your AI solutions?
8. Ethics: Who is accountable for the ethical considerations during each stage of the AI lifecycle?
9. AI talent: Does the organization know where AI talent exists and is it organized to benefit from AI discoveries?
10. Performance metrics: What are the current or future performance metrics most affected by insights derived from AI?

Data governance and model governance are vital for ethical AI development, ensuring compliance and quality. AI governance, on the other hand, addresses broader societal implications and organizational values. Together, these frameworks ensure responsible and effective AI utilization, fostering trust and aligning AI practices with ethical standards and societal expectations.

Read more stories about data ethics from SAS bloggers