AI Associate Practice Test 37 — 15 Questions

Question 1

A global company uses Salesforce Einstein Discovery to predict customer churn. They have a dataset with fields: Customer_Since__c (date), Last_Interaction_Date__c (date), Support_Cases__c (number), Product_Usage__c (percentage), Region__c (picklist), and Churned__c (boolean target). The model was trained and deployed, but predictions show bias against customers in the "EMEA" region. The data scientist notices that in the training data, 80% of EMEA customers are labeled as churned, while only 20% of other regions. Additionally, the Product_Usage__c field has many missing values for EMEA customers. The company wants to retrain the model to reduce bias. What is the best course of action?

Accepted Answer

Preprocess the data to impute missing Product_Usage__c values using region-specific averages, and then rebalance the dataset using stratified sampling. Option D is correct because it addresses both the missing data (impute using region-specific averages to preserve regional patterns) and the class imbalance (stratified sampling ensures balanced representation across regions during training). Option A removes Region, losing valuable information; Option B uses synthetic data which may introduce artificial patterns; Option C only rebalances but does not fix missing data which could still bias the model.

Answer

Oversample EMEA churned customers and undersample non-churned from other regions

Answer

Increase the sample size of EMEA customers by adding synthetic data

Answer

Remove the Region__c field from the model and retrain

Question 2

A company is building a chatbot using Einstein Bot's AI capabilities. They want to train intent recognition using historical chat transcripts. The transcripts contain many typos (e.g., 'hellp' instead of 'help') and slang (e.g., 'gonna' instead of 'going to'). The initial model performs poorly, misclassifying many intents. What data cleaning step is most important?

Accepted Answer

Normalize text by applying spell-correction and replacing slang with standard terms.. Normalizing text by correcting common typos and expanding slang reduces vocabulary sparsity and helps the model learn consistent word associations, improving intent recognition.

Answer

Use a spell-checker only for words that appear infrequently.

Answer

Keep the raw text as is because it reflects real user behavior.

Answer

Remove all messages that contain typos or slang to clean the dataset.

Question 3

A telecom company uses Einstein Discovery to predict customer churn. The training dataset contains 100,000 records, but only 5% represent churned customers. The model achieves 95% accuracy on a holdout test set, but the recall for churn is only 20%. The business wants to proactively retain at-risk customers, so they need to identify as many churners as possible. What action should the data scientist take to improve churn recall?

Accepted Answer

Oversample the minority class using SMOTE to create synthetic churn examples.. Class imbalance causes the model to favor the majority class. Oversampling the minority class (e.g., using SMOTE) balances the dataset, helping the model learn churn patterns better and improve recall.

Answer

Increase the regularization parameter to prevent overfitting.

Answer

Collect more data, especially of churned customers.

Answer

Undersample the majority class to match the minority class size.

Question 4

A healthcare provider implements Data Cloud to predict patient readmission rates. They have HIPAA compliance requirements. The data includes sensitive patient health information (PHI). The AI model must be trained without exposing PHI to unauthorized users. The data architect uses Data Cloud's data masking on PHI fields. However, model performance drops significantly after masking because the masked values lose predictive value. What additional step should the architect consider to maintain model performance while protecting PHI?

Accepted Answer

Use tokenization for highly predictive fields like diagnosis codes instead of masking. Option C is the best approach because tokenization preserves the relationship between values (e.g., diagnosis codes) while obscuring the actual PHI. This allows the model to learn patterns without exposing sensitive data. Option A violates HIPAA. Option B is not directly available in Einstein Studio as a built-in feature; differential privacy might be complex to implement. Option D does not address the masking issue.

Answer

Implement differential privacy within Einstein Studio

Answer

Remove masking and rely on user permissions to restrict access

Answer

Increase the volume of training data to compensate for masking

Question 5

A global e-commerce company deploys Einstein Bots in multiple countries. The bot uses natural language processing to handle customer returns. In one region, customers frequently complain that the bot does not understand their local dialect and incorrectly rejects valid returns. The company wants to maintain consistent customer experience while respecting regional diversity. The bot's language model was trained mainly on English data from the US and UK. The AI ethics board is concerned about fairness and transparency. They consider four options: (A) use a single, centrally-trained model with fallback to human agents for non-English queries, (B) deploy separate models fine-tuned on each dialect but with centralized monitoring, (C) disable the bot in regions with dialect issues, (D) use a translation layer to convert all inputs to English before processing. What is the best ethical approach?

Accepted Answer

Deploy separate models fine-tuned on each dialect with centralized monitoring. Option B is correct because fine-tuning on local dialects improves accuracy and fairness, while centralized monitoring ensures oversight. Option A is wrong: fallback to human agents is good but still may cause delays and dissatisfaction; the model itself is not inclusive. Option C is wrong: disabling the bot denies service and is not inclusive. Option D is wrong: translation may lose nuance and introduce errors.

Answer

Use a single centrally-trained model with fallback to human agents for non-English queries

Answer

Use a translation layer to convert all inputs to English before processing

Answer

Disable the bot in regions with dialect issues

Question 6

A government agency uses an AI system to allocate resources for public services such as healthcare and education. The system is designed to optimize overall efficiency based on historical usage data. After deployment, it becomes clear that underserved regions with less historical data receive significantly less funding than well-served regions. The agency's mission is to promote equity. The system's performance metrics show high efficiency, but community leaders protest the unfair distribution. What should the agency do?

Accepted Answer

Redesign the system to include fairness constraints that ensure minimum resource levels for underserved regions.. Option B is correct because incorporating fairness constraints ensures equitable distribution while still using AI to optimize. Option A ignores the fairness issue. Option C is good but may not be sufficient if the model still biases against underrepresented areas. Option D reverts to a less efficient system.

Answer

Abandon the AI system and use a manual, rule-based allocation system.

Answer

Collect more historical data from underserved regions before making adjustments.

Answer

Continue using the system as is, since it maximizes efficiency.

Question 7

A hospital uses an AI triage system to prioritize patients in the emergency department. The AI was trained on historical patient data and assigns priority scores based on vital signs and symptoms. Recently, a study finds that the system consistently assigns lower priority to elderly patients compared to younger patients with similar clinical presentations. The hospital's ethics committee is concerned about age discrimination. The current model achieves high accuracy in predicting outcomes, and doctors have come to rely on it for efficiency. What should the hospital do to address the ethical concern while maintaining clinical effectiveness?

Accepted Answer

Retrain the model with a modified objective that penalizes age-based disparities.. Option A is correct because adjusting the objective function to penalize age bias directly addresses the discrimination while keeping the model effective. Option B removes age, but bias may persist through correlated features. Option C requires building a new model from scratch, which is time-consuming and may not be necessary. Option D ignores the problem.

Answer

Replace the AI triage system with a completely new model built from scratch.

Answer

Continue using the current model since it has high accuracy and efficiency.

Answer

Remove age as an input feature from the model.

Question 8

A retail company uses Salesforce Einstein Vision to analyze customer images for product recommendations. The AI team notices that the model performs poorly on images of customers with darker skin tones, leading to fewer recommendations for that demographic. The team has access to a dataset of diverse skin tones but the company's data privacy policy prohibits using demographic data in training. What should the team do?

Accepted Answer

Document the bias, escalate to the ethics board, and seek guidance on using diverse data while maintaining privacy.. Option D is correct because it addresses the bias through proper channels while respecting privacy policies. Option A is wrong because removing labels may still encode bias and could violate the spirit of privacy. Option B is wrong because ignoring performance disparities is unethical. Option C is wrong because violating privacy regulations is not acceptable.

Answer

Retrain the model on the diverse dataset without considering privacy regulations.

Answer

Ignore the performance disparity as it is a result of natural data distribution.

Answer

Use the diverse dataset but remove skin tone labels to avoid privacy issues.

Question 9

An insurance company uses an AI model to set auto insurance premiums. The model uses factors including driving history, age, and ZIP code. A regulator finds that premiums in certain low-income neighborhoods are significantly higher than in affluent neighborhoods with similar risk profiles. The company's actuaries argue that the model is actuarially sound because it accurately predicts claims based on historical data. The company wants to comply with ethical guidelines and avoid legal action. Which action should they take?

Accepted Answer

Incorporate a fairness constraint that requires similar premiums for similar risk profiles regardless of ZIP code.. Option B is correct because introducing a fairness check ensures that similar risk levels result in similar premiums across neighborhoods, addressing ethical concerns without discarding valid risk factors. Option A ignores the issue. Option C removes a potentially relevant factor, but may reduce accuracy. Option D is a band-aid that doesn't fix underlying bias.

Answer

Defend the model based on its actuarial accuracy and historical claims data.

Answer

Cap premium increases in low-income neighborhoods at a fixed percentage.

Answer

Remove ZIP code from the model inputs entirely.

Question 10

A large financial institution uses Einstein Discovery to automate loan pre-approval decisions. The model was trained on ten years of historical data. After deployment, the compliance team finds that the approval rate for minority groups is 15% lower than the majority group, even after controlling for credit score and income. The data is balanced across groups. The model uses features like zip code, employment history, and debt-to-income ratio. The institution has a strict policy of fairness and non-discrimination. The AI team proposes three options: (1) remove zip code and employment history from the model, (2) add a fairness constraint to the model training, (3) lower the decision threshold for minority groups to balance approval rates. The compliance officer must choose the most ethical and effective course of action that aligns with Salesforce AI ethical guidelines. Which option should they choose?

Accepted Answer

Add a fairness constraint to the model training. Option B is correct because adding a fairness constraint directly addresses bias without arbitrary threshold changes (Option C) and while removing features (Option A) may not eliminate bias due to correlated features. Option A is wrong because zip code and employment history may be proxies; removing them could reduce predictive power without fully solving bias. Option C is wrong because it applies different standards to groups, which may be discriminatory and illegal. Option D is to continue using the model, which is unethical.

Answer

Lower the decision threshold for minority groups

Answer

Remove zip code and employment history from the model

Answer

Continue using the model as is, since data is balanced

Question 11

A credit scoring company develops an AI model that includes social media activity as a factor. The model awards higher scores to individuals with many online connections and consistent posting. Consumer advocates argue that this penalizes individuals with limited internet access or those who value privacy. The company defends the model, stating that it predicts creditworthiness better than traditional models. However, a regulatory body is investigating potential discrimination. The company wants to address ethical concerns without completely abandoning the model. Which approach is most appropriate?

Accepted Answer

Conduct a thorough analysis to determine whether social media activity is a legitimate, non-discriminatory predictor of creditworthiness.. Option B is correct because validating the relevance of social media data through rigorous analysis ensures that the factor is both fair and predictive. Option A abandons an innovative feature without evidence of harm. Option C may not be enough if the feature is irrelevant. Option D is too narrow.

Answer

Remove social media data from the model immediately.

Answer

Increase the weight of traditional factors like income and payment history.

Answer

Continue using the current model but offer an alternative traditional scoring option.

Question 12

A large e-commerce company uses Salesforce Einstein to recommend products to customers. The AI model is trained on purchase history, browsing behavior, and demographic data including age and gender. Recently, the company received complaints that the model seems to recommend lower-priced items to female customers and higher-priced items to male customers for the same product categories. The data science team confirms the model has a statistically significant difference in recommendation value by gender. The company's ethical AI policy requires fairness, transparency, and human oversight. The compliance team is considering several actions. Which action should the company take first?

Accepted Answer

Conduct a thorough bias audit to identify all sources of disparate impact. The correct answer is B because conducting a bias audit is the ethical first step to understand the root cause and scope of the bias before taking corrective action. Option A is wrong because removing gender from the model may not eliminate proxy effects. Option C is wrong because disabling the model without analysis loses business value and does not address how to fix it. Option D is wrong because increasing prices for females would introduce a new bias and is unethical.

Answer

Adjust the model to increase recommendation prices for female customers

Answer

Disable the recommendation system until the issue is resolved

Answer

Immediately remove gender from the training data and retrain the model

Question 13

A retail company uses an AI recommendation engine to suggest products to online shoppers. The engine uses past purchase history and browsing behavior. Recently, a customer advocacy group publishes a report showing that the engine recommends higher-priced products to customers in affluent zip codes and lower-priced products to customers in lower-income areas, even when both groups have similar browsing histories. The company's revenue has increased since implementing the engine, and marketing teams are pleased. However, the company wants to maintain a reputation for fairness. Which action should the company take?

Accepted Answer

Implement fairness constraints to ensure similar recommendation distributions across demographic groups.. Option B is correct because implementing fairness constraints ensures recommendations are not systematically skewed, while still allowing personalization. Option A prioritizes revenue over ethics. Option C is too drastic and would lose benefits. Option D reduces personalization and may not be effective.

Answer

Show only the most popular products to all customers regardless of browsing history.

Answer

Discontinue the AI recommendation system entirely.

Answer

Keep the current system as it maximizes revenue.

Question 14

A financial services firm deploys Einstein Prediction Builder to predict loan default risk. The model uses sensitive attributes like zip code and age. During testing, the model shows a disparate impact on minority neighborhoods. The compliance team requires explanation of individual predictions for regulatory audits. The data science team wants to use a complex deep learning model that is not interpretable. Which approach best balances performance and ethical responsibility?

Accepted Answer

Use a simpler, interpretable model (e.g., logistic regression) that may have slightly lower accuracy but ensures transparency and reduces bias.. Option B is correct because a simpler, interpretable model ensures transparency and reduces bias, aligning with ethical AI principles. Option A is wrong because post-hoc explanations may not be reliable or accepted by regulators. Option C is wrong because adjusting thresholds per group is discriminatory and illegal. Option D is wrong because using the model on a subset does not resolve the underlying bias or compliance requirement.

Answer

Use the complex model but provide post-hoc explanations like SHAP values to satisfy compliance.

Answer

Use the complex model but only for a subset of customers to limit exposure.

Answer

Use the complex model and hide the disparate impact by adjusting thresholds per group.

Question 15

A non-profit organization uses Salesforce AI to help prioritize grant applications. The AI scores applications based on historical funding decisions, project impact, and community need indicators. After deployment, staff notices that applications from rural areas consistently receive lower scores than those from urban areas, even when project quality is similar. The organization's mission is to serve underserved communities, including rural areas. The AI model is trained on historical data that favored larger, urban projects. The ethics committee is meeting to decide next steps. What is the most appropriate action to align ethical AI with the organization's mission?

Accepted Answer

Retrain the model with new criteria that prioritize rural impact. The correct answer is C because the model output contradicts the mission, so the model should be retrained with updated criteria that emphasize rural support. Option A is wrong because manually overruling is not scalable and may introduce bias. Option B is wrong because accepting the bias violates the mission. Option D is wrong because providing explanations does not fix the unfair scoring.

Answer

Accept the AI scores but provide staff training on bias

Answer

Manually overrule the AI scores for rural applicants

Answer

Continue using the model but add explainability reports