Before training an AI model, which activity best supports governance and risk mitigation?

Question

Options

• ASelecting the largest available dataset
• BPerforming an AI impact assessment
• CDeploying the model in a test environment
• DPublishing a transparency notice

• A
  3% (1)
• B
  86% (25)
• C
  7% (2)
• D
  3% (1)

Explanation

Performing an AI impact assessment (B) is the foundational governance activity that happens before training begins - it systematically identifies potential harms, biases, legal risks, and affected stakeholders, directly enabling risk mitigation decisions about whether and how to proceed.

Why the distractors fall short:

• A (largest dataset) - size alone doesn't address governance; bigger datasets can amplify bias and introduce more risk without proper evaluation.
• C (test environment deployment) - this is a post-training validation step, not a pre-training governance activity.
• D (transparency notice) - a transparency notice is a communication artifact, typically published after decisions are made, not a risk mitigation tool that shapes those decisions.

Memory tip: Think of an impact assessment as the "should we build this, and how?" question - it must come before you build (train). Everything else (testing, publishing, dataset selection) happens downstream of that governance decision.

No community discussion yet for this question.