Question: Discuss how human intuition can help during error analysis when a machine learning system is performing below human level. Use a speech-recognition example to support your explanation.

Sample answer: When a model is not yet as accurate as a person, error analysis can reveal what clues humans use that the model is missing. For example, if a speech system hears the sentence "I need a new seat by the window" and confuses "seat" with "suite," a person can use the surrounding words and the situation to see which transcript makes sense. That insight helps developers understand which contextual signals matter and adjust the model so it learns to use the same kind of context.

Key points:

• The model is performing below human level.
• Human judgment helps identify useful context.
• The model’s mistakes can be analyzed to find missing cues.
• A speech example should show how context resolves an ambiguity.
• The findings can guide changes to the learning system.

Rubric: A good response states that the model is underperforming relative to humans, explains that human intuition reveals the clues used to choose the correct answer, and applies the idea to a speech-recognition example involving context.