AI-900 Practice Question: Describe features of generative AI workloads on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Measuring the compute cost impact of adding generative AI to an application

  Why it's wrong here

  Measuring compute cost focuses on infrastructure spend, GPU hours, and operational expenses when adding generative AI to an application, which is a financial planning or FinOps activity. Impact assessment is not about organizational spend, but about analyzing potential risks like generating biased, harmful, or nonsensical content and identifying which users could be harmed by those failures. Cost metrics cannot reveal whether the application will treat different demographic groups equitably or whether safeguards such as content filters are needed.

• ✓

  Identifying potential harms, affected groups, and mitigation measures before deploying AI applications

  Why this is correct

  This is the essence of an AI impact assessment: a pre-deployment process used to systematically identify potential harms, the individuals or communities most likely to be affected, and the safeguards that can reduce those risks. In Microsoft's Responsible AI approach, this means documenting intended use, testing for failure modes such as inaccuracy or bias, and designing mitigations like guardrails, human oversight, and monitoring plans. Because the goal is to address harms before they reach users, this forward-looking evaluation is the correct definition.

• ✗

  Measuring user satisfaction scores after a generative AI feature launches

  Why it's wrong here

  Measuring user satisfaction scores after launch is a product analytics activity that captures user feedback, CSAT, or NPS, and it reflects whether users like the feature, not whether it caused harm. Impact assessment must be done before deployment because it requires changing design and controls; a post-hoc satisfaction survey cannot surface harms suffered by non-users, marginalized groups, or downstream systems. Satisfaction also fails to measure safety failures, content policy violations, or fairness gaps that the feature may have introduced.

• ✗

  Calculating the environmental impact of AI model training in terms of CO2 emissions

  Why it's wrong here

  Calculating CO2 emissions from AI training is a sustainability measurement that quantifies energy use, hardware efficiency, and environmental footprint. While environmental sustainability is one important dimension of responsible AI, an impact assessment specifically evaluates social, ethical, and safety harms such as bias, privacy breaches, and unintended consequences. This option therefore measures environmental impact rather than identifying affected groups or mitigation measures before deployment.

Go deeper

Related to this question