AI-900 Practice Question: Describe Artificial Intelligence workloads and considerations

Answer analysis

Option-by-option breakdown

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

• ✗

  Automatically generating descriptive captions for images using a pre-trained model

  Why it's wrong here

  Image captioning, typically performed by a pre-trained vision-language model, produces natural-language descriptions of image content. In contrast, AI-assisted labelling is designed to suggest structured annotation labels, such as object classes or bounding boxes, that a human must review and correct. While captioning is a valid computer vision task, its output is fundamentally different from the pre-populated, human-verified labels central to AI-assisted labelling.

• ✓

  Using a partially trained model to pre-populate labels that human annotators verify and correct

  Why this is correct

  This is exactly the human-in-the-loop workflow that defines AI-assisted labelling in Azure Machine Learning. A partially trained model proposes labels for new, unlabeled data, and human annotators verify and correct those suggestions. The corrected labels are then fed back into training, which improves the model and reduces the annotation burden over successive iterations.

• ✗

  Deploying a model to production without any human review of its outputs

  Why it's wrong here

  Deploying a model to production without any human review eliminates the verification step that is essential to AI-assisted labelling. In AI-assisted labelling, the human annotator always reviews and corrects the model's suggestions, creating a quality control loop that ensures label accuracy. Unreviewed deployment describes autonomous decision-making or full automation, which is a separate concept from the assisted annotation workflow and also bypasses the feedback necessary for iterative model improvement.

• ✗

  Using AI to detect and remove incorrectly labelled examples from a completed dataset

  Why it's wrong here

  This describes a post-hoc dataset validation step that uses anomaly detection or classification confidence to flag potentially mislabeled examples. AI-assisted labelling, however, is an interactive process that occurs during the annotation phase, pre-populating labels for the annotator to approve or fix in real time. Removing incorrect labels from a completed dataset focuses on cleaning data after labelling, whereas AI-assisted labelling prevents errors by involving humans earlier in the loop.

Go deeper

Related to this question