AI Associate Ethical Considerations of AI Practice Test 7 — 15 Questions

Question 1

A sales team uses Einstein Lead Scoring. They notice leads from certain industries are always low-scored. What should they do?

Accepted Answer

Review training data for bias. Low scores for specific industries often indicate bias in the training data, where historical lead data may have underrepresented or mislabeled those industries. Reviewing the training data for bias allows the team to identify and correct such imbalances, ensuring the Einstein Lead Scoring model produces fair and accurate predictions across all segments.

Answer

Retrain the model weekly

Answer

Ignore the scores

Answer

Use a different AI system

Question 2

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 3

Which TWO actions align with ethical AI practices in Salesforce?

Accepted Answer

Monitoring models for bias. Option C is correct because ethical AI in Salesforce requires ongoing bias detection and mitigation—Salesforce's Einstein Trust Layer and Model Cards emphasize regularly monitoring models for disparate impact across protected groups so that biased outcomes are caught and corrected. Option E is correct because providing explanations for AI predictions supports transparency and accountability, aligning with Salesforce's responsible AI principles and emerging regulations (e.g., GDPR Article 22) that give users the right to understand automated decisions. Option A is not inherently ethical—restricting training to internal data can still violate privacy or bias concerns, and ethical AI does not mandate internal-only data. Option B is wrong because fully automating decisions without human review removes the human oversight that ethical AI frameworks require for high-impact choices. Option D is wrong because using customer data without explicit consent violates privacy and consent principles central to ethical AI and Salesforce's data-use policies.

Answer

Using only internal data for training

Answer

Automating all decisions without human review

Answer

Using customer data without explicit consent

Question 4

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 5

Which THREE are key ethical considerations for AI according to Salesforce?

Accepted Answer

Accountability. According to Salesforce's guidance on ethical AI, Accountability (A) is a key consideration because organizations must be able to explain and take responsibility for their AI systems' decisions and outcomes, ensuring human oversight and clear ownership. Transparency (C) is also essential, as AI systems should be understandable and their decision-making processes open to scrutiny, so users and stakeholders can trust and audit them. Privacy (D) is a core ethical consideration because AI must protect personal data, respect consent, and comply with regulations, preventing misuse or unauthorized exposure of sensitive information. The unmarked options do not belong: Profitability (B) is a business objective rather than an ethical principle, and Speed (E) is a performance metric, not an ethical requirement for responsible AI.

Answer

Profitability

Answer

Speed

Question 6

Refer to the exhibit. A company uses this policy for a customer-facing AI model. What is the most critical ethical risk?

Accepted Answer

Lack of transparency. Lack of transparency. The policy explicitly states 'explainability: none', which means the AI model's decisions cannot be explained or understood by users. This lack of transparency prevents users from knowing why a decision was made, challenging the decision, or trusting the system, making it the most critical ethical risk. Option A (Performance) is not indicated in the policy and is less critical than transparency. Option B (Data privacy) is important but the policy does not mention any privacy violations. Option C (Model accuracy) is not addressed in the policy and while important, transparency is more fundamental for ethical AI.

Answer

Performance

Answer

Data privacy

Answer

Model accuracy

Question 7

Refer to the exhibit. What action should be taken?

Accepted Answer

Retrain the model with balanced training data. The correct action is D: retrain the model with balanced training data, because the exhibit shows a classification model whose performance metrics (such as precision, recall, or confusion matrix) indicate bias toward one class due to imbalanced training data. Retraining with balanced data addresses the root cause by giving the model representative examples of each class, improving fairness and overall predictive performance. Increasing the accuracy threshold (A) only changes the decision cutoff and does not fix the underlying data imbalance, while removing the model (B) discards a potentially salvageable asset, and deploying as is (C) would ship a biased or unreliable model into production.

Answer

Increase the accuracy threshold

Answer

Remove the model

Answer

Deploy as is

Question 8

Refer to the exhibit. Which ethical principle is violated?

Accepted Answer

Transparency. Transparency, because the scenario involves concealing information or failing to disclose how a system operates, which directly violates the ethical principle of transparency. Transparency requires that relevant stakeholders have visibility into data use, decision-making, and system behavior, and hiding such details undermines informed consent and trust. Safety (A) concerns preventing harm, fairness (B) concerns equitable treatment and bias, and accountability (C) concerns taking responsibility for actions, none of which are the primary violation when information is deliberately withheld.

Answer

Safety

Answer

Fairness

Answer

Accountability

Question 9

A company uses an AI model to screen job applications. They discover the model is less likely to recommend female candidates. What should the company prioritize first?

Accepted Answer

Implement a fairness metric to evaluate bias.. The company should first implement a fairness metric to evaluate bias (Option C), because before mitigating the problem it must quantify and measure the disparity in the model's recommendations across genders using metrics such as demographic parity or equal opportunity. Without a defined fairness metric, any remediation is unguided and cannot be validated. Removing gender-related features (Option A) is a possible later mitigation, but proxy variables can still encode gender, so it is not the first priority. Retraining with only male candidates (Option B) would worsen and entrench the bias, and simply increasing training data size (Option D) does not guarantee reduced bias and may even amplify existing imbalances.

Answer

Remove gender-related features from the model.

Answer

Re-train the model with only male candidates.

Answer

Increase the model's training data size.

Question 10

A Salesforce admin wants to deploy an Einstein bot that uses natural language processing. Which practice best ensures ethical use?

Accepted Answer

Provide clear disclaimers that the user is interacting with an AI.. Providing clear disclaimers that the user is interacting with an AI is a fundamental ethical practice for transparency, ensuring users are aware they are not communicating with a human. This aligns with ethical AI principles such as honesty and informed interaction, which are critical for Einstein bots using NLP. It helps set appropriate expectations and builds trust.

Answer

Use the bot only for internal processes.

Answer

Collect as much personal data as possible to improve accuracy.

Answer

Allow the bot to make autonomous decisions without human review.

Question 11

A data scientist is training a model to predict customer churn. To ensure fairness, what should the data scientist do?

Accepted Answer

Ensure the training data is representative of the entire customer base.. Ensuring the training data is representative of the entire customer base directly addresses fairness by preventing underrepresentation or overrepresentation of specific demographic groups. A representative dataset helps the model learn unbiased patterns across all segments, reducing the risk of disparate impact. This aligns with the principle of fairness in AI, where the model's predictions should not systematically disadvantage any group.

Answer

Focus solely on model accuracy ignoring demographic groups.

Answer

Remove all demographic attributes from the dataset.

Answer

Use only historical data without checking for bias.

Question 12

A company deploys an AI system that recommends loan amounts. They want to ensure explainability. Which approach best aligns with ethical AI?

Accepted Answer

Provide loan recommendations with a detailed rationale.. Providing loan recommendations with a detailed rationale directly supports explainability, allowing stakeholders to understand how and why a specific loan amount was recommended, which is a core ethical AI principle for high-stakes decisions like lending. Explainability in this scenario means the system's outputs can be interpreted and justified, helping ensure fairness, accountability, and regulatory compliance. Option A does not fit because dynamically updating features without documentation reduces transparency and traceability. Option B does not fit because limiting use to internal employees does not make the system's decisions explainable. Option D does not fit because a black-box neural network prioritizes accuracy over interpretability, which conflicts with the requirement for explainability.

Answer

Allow the system to update its features dynamically without documentation.

Answer

Restrict the system's use to only internal employees.

Answer

Use a black-box neural network for highest accuracy.

Question 13

An organization uses Einstein to predict sales opportunities. They notice the model performs poorly for small businesses. What is the most ethical approach?

Accepted Answer

Retrain the model with more data from small businesses.. Retrain the model with more data from small businesses. Poor performance for small businesses most likely reflects underrepresentation in the training data, so adding more small-business examples improves the model's accuracy and fairness for that segment without abandoning it. This is the most ethical approach because it directly addresses the bias rather than hiding or accepting it. Option A does not fix the underlying data imbalance and may reduce overall accuracy, option B creates a fragmented solution that can still be biased and harder to govern, and option D is unethical because it ignores a customer segment based on revenue rather than addressing the model's deficiency.

Answer

Reduce the model's complexity to improve performance.

Answer

Use a completely different model for small businesses.

Answer

Ignore small businesses as they contribute little revenue.

Question 14

A team is developing a chatbot for customer service. To ensure ethical AI, which practice should be incorporated?

Accepted Answer

Allow the chatbot to escalate to a human agent upon request.. Human escalation supports accountability and user control. Option B is wrong because indefinite data storage violates privacy. Option C is wrong because a single data source may introduce bias. Option D is wrong because mimicking emotions may deceive users.

Answer

Store all conversation data indefinitely for analysis.

Answer

Use a single data source for training to avoid inconsistency.

Answer

Design the chatbot to mimic human emotions perfectly.

Question 15

A financial institution uses an AI model to approve credit. The model shows disparate impact against a protected group. Under Salesforce's ethical AI principles, what is the most appropriate action?

Accepted Answer

Apply a bias mitigation technique such as reweighing.. Salesforce's ethical AI principles emphasize responsible use and mitigating harmful bias, so applying a bias mitigation technique such as reweighing directly addresses the disparate impact while preserving the model's utility. Reweighing adjusts sample weights during training to reduce bias against the protected group, which is a targeted, proportionate remedy for a model that is otherwise functional. Discontinuing the model immediately (A) is an overreaction that discards a viable system instead of remediating it, and increasing the decision threshold for all applicants (B) does not correct group-specific bias and may worsen outcomes. Publishing predictions publicly (C) exposes sensitive financial data and does not itself mitigate the bias, so it is not the appropriate action.

Answer

Discontinue the model immediately.

Answer

Increase the model's decision threshold for all applicants.

Answer

Publish the model's predictions publicly for scrutiny.