Case context: A team's machine learning system for approving loan applications has already surpassed the accuracy of human loan officers. The team is now finding it much harder to make further improvements than they did earlier in the project.

Question: Using the concept from the source, diagnose why this team's progress has slowed and what this implies about the techniques still available to them.

Sample answer: The team's progress has slowed because their algorithm now surpasses human-level performance, meaning humans can no longer easily identify examples where the algorithm is clearly wrong. As a result, only a subset of human-comparison techniques still apply, which naturally reduces the pace of further improvement compared to when the algorithm was still catching up to human performance.

Key points:

• Loan approval is cited in the source as an application where machines surpass human-level performance
• Progress slows once machines surpass human-level performance
• Humans struggle to identify clearly wrong examples once machines surpass their performance
• Only a subset of human-comparison techniques remain applicable

Rubric: Full credit identifies that surpassing human-level performance is the cause, references loans as an example from the source of a domain where machines surpass humans, and explains the reduced applicability of human-comparison techniques.