Understanding How to Govern AI Development

Governance responsibilities for designing, building, training, testing, releasing, monitoring and maintaining AI models and systems throughout the development life cycle.

5 minutes 5 Questions

Understanding How to Govern AI Development is a critical competency for AI Governance Professionals, encompassing the frameworks, policies, and practices needed to ensure AI systems are developed responsibly, ethically, and in compliance with applicable regulations. At its core, governing AI devel…

Concepts covered
Defining AI System Business Context and Use CaseDocumenting the AI Design and Build ProcessData Quality, Quantity and Integrity for AI TrainingUnit, Integration and Validation Testing for AIManaging Issues and Risks During AI Training and TestingConformity Assessment Requirements for AI ReleaseAI Maintenance, Updates and Retraining ScheduleSecurity Testing for AI SystemsModel and Data Drift in Production AIPublic Disclosures and Transparency Obligations for AIAI Impact Assessment in DesignRequirements Gathering for AI SystemsAI Architecture and Model SelectionHuman Oversight in AI DesignMetric and Threshold Evaluation for AIStakeholder Engagement and Feedback in AI DesignOperational Controls During AI DevelopmentProbability and Severity Harms Matrix for AI RiskRisk Mitigation Hierarchy for AIStakeholder Mapping for AI RiskBenchmarking and Pre-Deployment Pilots for AIData Governance for AI Training: Rights and Fit-for-PurposeData Lineage and Provenance for AIPerformance, Security, Bias and Interpretability TestingDocumenting the AI Training and Testing ProcessRelease Readiness Assessment and Model CardsContinuous Monitoring of Production AI SystemsAudits, Red Teaming and Threat Modeling for AIAI Incident Management and Root-Cause AnalysisPost-Market Monitoring Plans for AIInstructions for Use Provided to AI Deployers
Test mode:
Start practice test

AIGP - Understanding How to Govern AI Development Example Questions

Test your knowledge of Understanding How to Govern AI Development

Question 1

A team is starting to gather requirements for a new AI chatbot that will handle customer service inquiries. As the governance professional, what should be the first step in the requirements gathering process?

Correct Answer: Identify all relevant stakeholders who will be affected by or interact with the AI system

Identifying all relevant stakeholders is the essential first step in requirements gathering because understanding who will be affected by, interact with, or have authority over the AI system ensures that all perspectives, needs, risks, and regulatory considerations are captured before any design or development decisions are made.

Defining technical architecture is premature without first understanding stakeholder needs and constraints. Collecting training data should follow after requirements are clearly defined through stakeholder input. Establishing performance benchmarks also comes later, as meaningful metrics can only be set once stakeholder expectations and use cases are understood.

The key concept tested is stakeholder identification as the foundational step in responsible AI requirements gathering and governance.

Question 2

A team is building an AI chatbot to answer simple customer FAQs from a fixed list of 50 questions. The answers are predefined and do not change. The team needs to select the most appropriate AI approach. Which model selection strategy is best suited for this use case?

Correct Answer: Use a rule-based or keyword-matching system since the task involves fixed, predefined responses

A rule-based or keyword-matching system is the most appropriate choice because the task involves a small, fixed set of 50 predefined question-answer pairs with no need for dynamic or generative responses, making simplicity, transparency, and efficiency the guiding principles.

Training a large language model from scratch is massively disproportionate to the task's simplicity and scale. Deep reinforcement learning is designed for sequential decision-making and is unnecessary when answers are static and predefined. Fine-tuning a generative transformer introduces unnecessary complexity, cost, and risk of hallucination when responses should be exact and unchanging.

The key concept tested is proportionality in model selection, matching AI complexity to task requirements.

Question 3

A team is deploying an AI model to classify loan applications as approved or denied. They need to select a performance metric and set an acceptable threshold. The model will be used in a regulated financial environment where fairness is critical. Which approach should the team take when setting the evaluation threshold for this model?

Correct Answer: Set the threshold based on balancing false positive and false negative rates while considering fairness across demographic groups

In a regulated financial environment where fairness is critical, the correct approach is to balance false positive and false negative rates while explicitly considering fairness across demographic groups, because both types of errors have significant consequences (wrongly denying qualified applicants or approving unqualified ones) and disparate impact across protected classes can violate fair lending laws.

Maximizing overall accuracy can mask significant disparities in error rates across demographic groups, perpetuating discrimination. Using precision alone ignores false negatives, meaning qualified applicants from certain groups could be systematically denied. A default 0.5 threshold is arbitrary and not tailored to the specific risk profile or fairness requirements of the use case.

The key concept tested is fairness-aware threshold setting in high-stakes, regulated AI applications.

More Understanding How to Govern AI Development questions
834 questions (total)
Practice 100 questions
One session at a time, always new questions