AI Associate Ethical AI and Data Privacy Practice Question
A data scientist is building a churn prediction model using Einstein Discovery. They want to ensure the model does not rely on sensitive attributes like race or gender, even if those are correlated with other features. Which technique is MOST aligned with Salesforce's data minimisation principle?
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Remove sensitive attributes from the training data and avoid using proxies that strongly correlate with them
Data minimisation means using only the data necessary for the task. Excluding sensitive attributes from the feature set is the most direct way to prevent them from being used, even if they are correlated with other features. Correlation does not imply causation, and if those features are not essential, they should be removed.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Remove sensitive attributes from the training data and avoid using proxies that strongly correlate with them
Why this is correct
Removing sensitive attributes directly and also being cautious of proxy features minimizes the chance of the model indirectly using protected characteristics.
- ✗
Include all features and rely on the AI to ignore biased ones
Why it's wrong here
Models can learn correlations with sensitive attributes even if they are not directly included; simply including them risks encoding bias.
- ✗
Apply a fairness constraint after training to adjust predictions
Why it's wrong here
Post-hoc adjustments can help but are less principled than minimizing data use from the start. Data minimisation is proactive.
- ✗
Use differential privacy to add noise to the training data
Why it's wrong here
Differential privacy protects individual records but does not address the inclusion of sensitive features; it is about privacy, not fairness.
Go deeper
Related to this question
About these practice questions
One of 753 original AI Associate practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. Learn why practice questions differ from exam dumps →
Same concept, more angles
4 more ways this is tested on AI Associate
These questions test the same concept from different angles. Work through them to make sure you can recognise it however the exam phrases it.
Variation 1. A company wants to use Einstein Prediction Builder to predict which customers are likely to churn. They have a dataset that includes customers' names, email addresses, and detailed purchase history. According to the data minimisation principle, which fields should be included in the model?
easy- ✓ A.Only the purchase history
- B.Only the email addresses, since they are unique identifiers
- C.All three fields because more data improves accuracy
- D.Names and purchase history, but not email addresses
Why A: Data minimisation means using only the data necessary for the prediction. Purchase history is relevant to churn prediction, while names and email addresses are personally identifiable information (PII) that are not needed for the model and should be excluded to minimize privacy risks.
Variation 2. A financial services company is deploying Einstein Prediction Builder to predict loan default risk. They are concerned about using sensitive attributes like race or gender in the model. Which data governance practice should they apply?
medium- A.Use synthetic data to replace sensitive attributes with random values.
- B.Include all available attributes to maximize model accuracy, then apply fairness constraints.
- C.Mask the sensitive attributes but still include them in the model training.
- ✓ D.Exclude sensitive attributes from the model features unless they are essential and legally permitted, and ensure no proxies exist.
Why D: Data minimisation is a core principle: only use relevant features for the prediction. Sensitive attributes that could lead to discriminatory decisions should be excluded unless legally required and properly managed.
Variation 3. An organization using Einstein Prediction Builder wants to ensure that no customer personally identifiable information (PII) is used in model training. Which data governance practice should they enforce?
hard- A.Enabling zero data retention in the Trust Layer
- B.Data anonymization via the Einstein Trust Layer
- C.Regularly auditing the model for bias
- ✓ D.Data minimisation by selecting only non-PII fields as predictors
Why D: The question specifically asks how to ensure no PII is used in model training. Data minimization by selecting only non-PII fields as predictors directly prevents PII from entering the training dataset at the source. This is a proactive governance practice that avoids reliance on post-processing or masking, which may still expose PII during intermediate steps.
Variation 4. A healthcare organization uses Einstein Prediction Builder to predict patient no-show rates. They want to ensure that protected health information (PHI) like patient names and social security numbers are not used in the model. Which Salesforce Trusted AI principle or feature directly addresses this requirement?
medium- A.Zero data retention
- B.Explainability
- C.Human oversight
- ✓ D.Data minimisation
Why D: Data minimisation is a key principle: only use relevant features for the model and avoid including sensitive PII unnecessarily. The Einstein Trust Layer's PII masking also helps, but the principle that directly addresses using only relevant fields is data minimisation.
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This AI Associate practice question is part of Courseiva's free Salesforce certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AI Associate exam.