AI Associate Practice Test 15 — 15 Questions

Question 1

A financial services company uses Einstein AI to recommend credit limits. The model tends to assign lower limits to applicants from a certain region. Which action best aligns with ethical AI practices?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Investigate the data and model for bias, and adjust the model if necessary. Option B is correct because ethical AI practice requires investigating potential bias in both the training data and the model's outputs, then mitigating it—such as through data rebalancing, fairness constraints, or retraining—so the model does not systematically disadvantage applicants from a particular region. This aligns with principles of fairness, accountability, and transparency in AI governance. Option A does not fit because discarding the AI for a rule-based system avoids rather than addresses the underlying bias and may sacrifice legitimate model benefits. Option C is wrong because manually inflating limits for an entire region is itself a discriminatory, non-individualized fix that ignores actual creditworthiness. Option D is wrong because statistical validity does not make a biased or discriminatory outcome ethically acceptable.

Answer

Replace the AI model with a simpler rule-based system

Answer

Manually increase credit limits for all applicants from that region

Answer

Ignore the pattern since the model is statistically valid

Question 2

Which THREE are key principles of Salesforce's AI Ethics framework? (Choose three.)

Accepted Answer

Be Inclusive. Options A, C, and D are correct. Option B is wrong because 'speed over accuracy' contradicts responsible AI. Option E is wrong because 'maximum profit' is not an ethical principle.

Answer

Be Fast Over Accurate

Answer

Be Profit-Maximizing

Question 3

An organization wants to implement AI in a way that builds trust. Which practice is most important?

Accepted Answer

Providing explanations for AI decisions.. Providing explanations for AI decisions increases transparency and trust. Option A is wrong because secrecy reduces trust. Option C is wrong because complexity can reduce trust if not explainable. Option D is wrong because cheap data sources may compromise data quality and ethics.

Answer

Keeping model details secret.

Answer

Using complex models for better performance.

Answer

Using the cheapest data sources.

Question 4

An organization uses Einstein Recommendation Builder to suggest products. They want to ensure recommendations are fair across demographics. Which action should they take?

Accepted Answer

Regular bias audits. Regular bias audits (option C) are the right action because fairness across demographics requires ongoing measurement and monitoring of recommendation outcomes for disparate impact, not a one-time configuration change. Auditing lets the organization detect and correct skewed product suggestions before they harm specific groups. Option A is wrong because using only demographic features would amplify rather than reduce bias. Option B is wrong because removing all user data would destroy the personalization signal Einstein Recommendation Builder needs and does not by itself guarantee fairness. Option D is wrong because increasing model depth adds capacity and complexity but does not address or measure demographic fairness.

Answer

Use only demographic features

Answer

Remove all user data

Answer

Increase model depth

Question 5

An AI system for medical diagnosis is trained on data from one region. When deployed globally, it performs poorly. This is an issue of?

Accepted Answer

Generalizability. Generalizability, because the model was trained on data from a single region and fails to perform well when applied to other regions, meaning its learned patterns do not transfer to new populations or distributions. This is a classic generalization gap, where performance degrades on out-of-distribution data not represented in the training set. Overfitting (B) is related but refers specifically to fitting training data too closely, including noise, rather than the broader failure to transfer across regions. Accountability (A) concerns responsibility and governance for system outcomes, and privacy violation (C) concerns improper use or disclosure of personal data, neither of which describes the cross-region performance drop.

Answer

Accountability

Answer

Overfitting

Answer

Privacy violation

Question 6

A health app collects users' location data for AI-driven recommendations, but users are not informed about this data collection. Which ethical principle is most compromised?

Accepted Answer

Data minimization and consent. Data minimization and consent, because the scenario describes users being unaware that their location data is collected and used for AI recommendations, which directly violates the requirement to obtain informed consent and to limit collection to what is necessary and disclosed. Transparency (A) is related but is a broader principle about openness and disclosure, whereas the specific ethical breach here centers on collecting and using personal data without user permission. Fairness (C) concerns equitable treatment and avoiding bias in outcomes, which is not the primary issue described. Accountability (D) concerns being answerable for data practices, but the core violation is the lack of consent and unnecessary data collection, not the absence of a responsible party.

Answer

Transparency

Answer

Fairness

Answer

Accountability

Question 7

A company wants to ensure their AI is fair. Which TWO steps are appropriate?

Accepted Answer

Test model performance on different demographic groups. Option D is correct because testing model performance on different demographic groups (e.g., measuring accuracy, false positive/negative rates across subgroups) is the core method for detecting disparate impact and bias in an AI system. Option E is correct because involving diverse stakeholders in model development brings varied perspectives that help surface hidden assumptions, define context-appropriate fairness criteria, and reduce blind spots that a homogeneous team would miss. Option A is not appropriate because relying on a single fairness metric is insufficient — fairness metrics such as demographic parity, equalized odds, and equal opportunity often conflict, so multiple metrics must be examined together. Option B is not appropriate because deploying quickly to gather real-world data prioritizes speed over safety and can cause harm to users before bias is identified and mitigated. Option C is not appropriate because simply removing sensitive attributes does not eliminate bias, since proxy variables (e.g., zip code, name) can still encode protected characteristics and cause discriminatory outcomes.

Answer

Use a single fairness metric to evaluate the model

Answer

Deploy the model quickly to gather real-world data

Answer

Remove all sensitive attributes from the data

Question 8

A company is deploying an AI system that makes recommendations to users. To ensure ethical use, they should:

Accepted Answer

Allow users to opt out and understand how decisions are made.. Allowing users to opt out and understanding decisions respects autonomy and transparency. Option B is wrong because making recommendations without oversight can lead to harmful outcomes. Option C is wrong because maximizing engagement without regard to user well-being is unethical. Option D is wrong because using only internal data may not be sufficient and could raise privacy concerns.

Answer

Make recommendations without oversight.

Answer

Maximize engagement regardless of user well-being.

Answer

Use only internal data.

Question 9

A Salesforce admin builds an Einstein Prediction Builder model to predict customer churn. The model assigns higher churn risk to customers in a certain demographic group. What is the MOST ethical FIRST step?

Accepted Answer

Analyze the data and model for potential bias. The most ethical first step is to analyze the data and model for potential bias (option C), because before taking any corrective action you must determine whether the disparity stems from biased training data, a proxy variable, or a legitimate business factor, and quantify the impact using fairness metrics. Only after this diagnostic assessment can you decide on an appropriate mitigation. Disabling the model immediately (option A) is premature and may discard a valid model without understanding the root cause. Removing demographic features (option B) is a possible mitigation but is not the first step, and it can still leave bias through correlated proxy variables. Retraining with more data (option D) does not address bias and may even amplify it if the additional data reflects the same historical inequities.

Answer

Disable the model immediately

Answer

Remove demographic features from the model

Answer

Retrain the model with more data

Question 10

Which TWO actions are most effective for ensuring fairness in an AI model used for loan approvals?

Accepted Answer

Regularly audit model outcomes for disparate impact across demographic groups.. Option A is correct because regularly auditing model outcomes for disparate impact across demographic groups is the core mechanism for detecting and quantifying bias in loan-approval decisions, using metrics such as the disparate impact ratio (e.g., the 80% rule) to verify that protected classes are not unfairly disadvantaged. Option D is correct because using diverse and representative training data ensures the model learns patterns that generalize fairly across the population it serves; biased or unrepresentative data is a primary root cause of unfair lending outcomes, so representative sampling directly reduces that risk. The remaining options do not belong: relying solely on historical data (B) perpetuates existing biases rather than correcting them, letting the model self-correct without human intervention (C) provides no guarantee of fairness and can drift further, and using a single simple algorithm (E) does not by itself ensure fairness since a simple model can still encode biased data or features.

Answer

Rely solely on historical data without any adjustments.

Answer

Allow the model to self-correct over time without human intervention.

Answer

Use a single, simple algorithm to avoid complexity.

Question 11

A Salesforce admin is configuring Einstein Search for an organization with users in multiple countries. Which ethical consideration is most important?

Accepted Answer

Language bias in the search results.. When Einstein Search serves users across multiple countries, the most important ethical consideration is language bias: search relevance models trained predominantly on one language or locale can return poorer or skewed results for other languages, effectively disadvantaging non-dominant-language users. This is a fairness and inclusion issue because search results influence what information users can find and act on. Ensuring balanced multilingual training data and evaluating relevance per locale is the ethical priority.

Answer

Stop word exclusion.

Answer

Query performance optimization.

Answer

Tokenization settings.

Question 12

A company is preparing customer data to train a custom AI model for sentiment analysis. Which two data preparation best practices should they follow? (Choose two.)

Accepted Answer

Ensure data is representative of all customer demographics.. Option B is correct because training data must be representative of all customer demographics to avoid sampling bias, which would cause the sentiment analysis model to perform poorly or unfairly for underrepresented groups. Option E is correct because anonymizing personally identifiable information (PII) before training satisfies data privacy and compliance requirements (e.g., GDPR, CCPA) and reduces the risk of leaking sensitive customer data into the model or its outputs. Option A is not a best practice because restricting data to only the last month introduces recency bias and may not capture the full range of sentiment expressions or seasonal patterns. Option C is wrong because dropping all records with missing values can discard valuable data and further skew the dataset; imputation or careful handling is usually preferred. Option D is wrong because labeling by a single annotator introduces individual bias and inconsistency; multiple annotators with inter-annotator agreement checks yield more reliable labels.

Answer

Use only data from the last month.

Answer

Remove all records with missing values.

Answer

Label data manually by a single annotator.

Question 13

Refer to the exhibit. What is the most likely cause of the fairness issue?

Accepted Answer

The training data is imbalanced, causing the model to perform better on the majority group.. The fairness issue is most likely caused by imbalanced training data, where one demographic group (e.g., males) is overrepresented, causing the model to learn patterns that favor the majority group and perform worse on the minority group. This is the classic source of disparate impact in ML fairness — the model reflects the statistical distribution of the data it was trained on, so underrepresentation of females leads to higher error rates for that group.

Answer

The model overfits to the male group.

Answer

The overall accuracy is too low.

Answer

The model is inherently biased against females.

Question 14

A data scientist notices that an Einstein model for predicting customer churn has unusually high accuracy on training data but performs poorly on validation data. Which data issue is the most likely cause?

Accepted Answer

A field containing future information (e.g., 'churn_date') was included in features. Including a field like 'churn_date' in the feature set introduces target leakage, where the model has access to information that would not be available at prediction time. This causes the model to appear highly accurate on training data (since it can directly 'see' the outcome) but fails to generalize to validation data where such future information is absent. In Salesforce Einstein, features must be strictly historical or static to avoid this data leakage issue.

Answer

The dataset has an imbalanced class distribution

Answer

The dataset contains many missing values

Answer

The model was trained on stale data from a different season

Question 15

Which TWO features are part of Einstein AI capabilities in Salesforce Sales Cloud?

Accepted Answer

Einstein Opportunity Scoring. Einstein Opportunity Scoring (A) is a Sales Cloud Einstein feature that uses AI to analyze opportunity field history, activity, and related records to assign each open opportunity a score from 1 to 100 indicating likelihood of closing, helping reps prioritize their pipeline. Einstein Lead Scoring (C) is likewise a Sales Cloud Einstein feature that evaluates lead fields and behavioral data to predict which leads are most likely to convert, displayed as a score on the lead record. Both are core predictive-scoring capabilities delivered within Sales Cloud Einstein. By contrast, Einstein Case Classification (B) and Einstein Article Recommendations (E) are Service Cloud features used for case field prediction and knowledge article suggestion, and Einstein Bots (D) is a Service Cloud conversational AI tool for customer self-service, so none of those belong to Sales Cloud capabilities.

Answer

Einstein Case Classification

Answer

Einstein Bots

Answer

Einstein Article Recommendations