Exercise: Module 6, Lesson 2 - Verification vs. Validation Failure Analysis
Objective: To develop a deep, practical understanding of the critical difference between Verification ("Did we build the thing right?") and Validation ("Did we build the right thing?") by analyzing real-world AI project failure scenarios.
Your Task
Below are three scenarios describing AI projects that have failed in some way. For each scenario, you must:
- Diagnose the Failure: State whether the primary failure was one of Verification or Validation.
- Justify Your Diagnosis: Write a brief (1-2 sentence) explanation for your choice, referencing the core definitions of verification and validation.
- Propose a Preventative Test: For Scenario 3 only, describe a specific test (either a verification or a validation test) that, if performed properly, could have caught this failure before the project was launched.
The Scenarios
Scenario 1: The Resume Screener
A company builds an AI tool to automatically screen resumes and rank the best candidates for engineering jobs. The technical requirement was to "identify candidates with the strongest technical skills based on our historical hiring data." The model is trained on 10 years of the company's hiring data. The system works perfectly according to the spec, but it is discovered that the model consistently ranks male candidates higher than female candidates, even with identical qualifications. This is because the historical data it was trained on reflected a past hiring bias.
Scenario 2: The Crop-Watering Robot
An agricultural tech company builds a robot that uses an AI model to predict the precise amount of water each individual plant needs. A technical requirement is that the robot must be able to operate for 12 hours on a single battery charge. During field testing, however, the robot consistently runs out of power after just 8 hours, leaving half the crops unwatered. The AI model for predicting water needs works perfectly, but the hardware does not meet its required specification.
Scenario 3: The AI-Powered News App
A startup launches a news app that uses a sophisticated AI to create a perfectly personalized news feed for every user, showing them only the stories they are most likely to engage with. The system is a technical marvel: it achieves a 95% click-through rate on its recommended articles, meeting its primary technical requirement. However, after several months, it becomes clear that the app is creating extreme "filter bubbles," where users are only shown content that confirms their existing biases, leading to increased polarization and user complaints about missing major news events.
Deliverable
Submit a Markdown file with your analysis for all three scenarios.
Example Submission Snippet:
Scenario 1: The Resume Screener
- Diagnosis: [Verification or Validation]
- Justification: [Your 1-2 sentence explanation]
Scenario 2: The Crop-Watering Robot
- Diagnosis: [Verification or Validation]
- Justification: [Your 1-2 sentence explanation]
Scenario 3: The AI-Powered News App
- Diagnosis: [Verification or Validation]
- Justification: [Your 1-2 sentence explanation]
- Preventative Test: [Your description of the test]