Get Certified Easily with Online RAI Dumps

Category:

Comments:

0 Comments

Post Date:

September 27, 2025

Using Risk and AI (RAI) RAI questions can help you build confidence as you prepare for the exam. By practicing with these RAI dumps questions, you will be better prepared and more confident when you take the actual exam. Besides, RAI exam dumps questions cover a wide range of topics that are relevant to the certification exam. By using RAI dumps questions, you will be able to ensure that you have a comprehensive understanding of the material covered on the exam. Practice free GARP RAI exam dumps below.

Page 1 of 10

1. A financial institution is using alternative data in its AI models to improve credit risk assessments.

What is an important action the institution should take to manage the risks associated with this data?

Allow data providers to modify data definitions without notification.

Accept minor inconsistencies as these will not affect the model's performance.

Regularly monitor the input data to identify changes in definitions, quality issues, and potential biases.

Use the alternative data only if it's publicly available.

2. A risk manager at a digital technology firm is considering ethical principles similar to the Hippocratic Oath for AI systems.

What should be a key ethical principle for AI development?

Maximizing user data collection for predictive accuracy

Ensuring AI systems do not harm users or society

Developing AI primarily for efficiency, without considering ethics

Allowing AI algorithms to operate without human oversight

3. How might AI exacerbate global inequality at a societal level?

By allowing less industrialized countries to adopt AI at a faster rate than industrialized countries

By increasing job opportunities equally across all countries

By benefiting wealthier, more industrialized countries more than others

By promoting open access to AI resources for all nations

4. A hedge fund manager wants to classify market sentiment from daily news headlines using a bag-of- words (BoW) approach. They choose a vocabulary consisting of specific finance-related words like “bull,” “bear,” “rally,” and “crash.”

How will this choice impact the BoW vectors for each headline?

Only words in the vocabulary will be counted in each headline’s BoW vector

Each headline will be analyzed for all unique words present in the corpus

The vocabulary will automatically update as new words appear

Each vector will contain sentiment scores for words in the vocabulary

5. An e-commerce platform wants to create a more compact representation of its product features. The team is considering using an autoencoder with fewer hidden units than input features.

What advantage does this setup provide?

Increased training time

Better generalization by reducing overfitting

Exact reconstruction of inputs

Higher accuracy in predictions

6. An analyst uses stepwise regression to build a predictive model for house prices. After starting with an empty model, the analyst adds features one by one until no further reduction in the Akaike Information Criterion (AIC) is achieved.

Which method is the analyst using?

Backward stepwise selection

Forward stepwise selection

LASSO regularization

Ridge regression

7. A financial services company is designing an AI model for personalized investment advice.

Which practice best demonstrates an ethical commitment beyond legal compliance?

Relying solely on legal regulations to guide AI model design.

Ensuring the model follows data protection laws like GDP

Including ethical considerations, like fairness and transparency, in the AI model’s decision-making process.

Limiting the AI’s ethical commitments to what is mandated by industry standards.

8. A financial institution uses an opaque consumer scoring algorithm.

What is a key reason for keeping the algorithm confidential?

Ensuring users can easily understand their scores

Protecting proprietary information and competitive advantage

Allowing competitors to improve similar systems

Sharing all code with the public for transparency

9. A risk analyst at a hospital is reviewing an AI tool for triaging emergency cases, prioritizing those with the most severe needs.

From a consequentialist perspective, which approach would be ethically justified?

Approve the tool if it maximizes survival rates and optimizes care resources

Reject the tool if it cannot guarantee 100% accuracy for each case

Approve the tool only if all stakeholders, including patients, fully consent

Reject the tool if any individual’s treatment might be delayed

10. To ensure effective model-risk management, a firm wants to establish model-risk tiers.

Why is this important?

It allows all models to be validated at the same frequency.

It reduces the need for defining validation expectations.

It helps tailor validation frequency and detail based on the model's risk tier.

It limits validation to only high-risk models.

Page 2 of 10

11. A credit scoring model using a neural network has 200 features in the input layer and a single neuron in the output layer.

To avoid overfitting, what network structure is likely to be most effective based on current best practices?

A "pyramid" structure with each successive layer having fewer neurons

A uniform structure with regularization

A structure with only one hidden layer of 200 neurons

Increasing the number of neurons in each hidden layer gradually

12. A retail company uses AI for personalized advertising but receives complaints about invasive targeting.

What should the company do to address potential reputational damage?

Continue using the AI without change

Provide customers with clearer options to manage their ad preferences

Disable the AI for all targeted ads

Collect even more customer data for better targeting

13. An insurance company wants to enhance its claims processing efficiency by using GenAI to automate customer interviews for claims.

Which GenAI application aligns best with this goal?

Using LLMs for monitoring social media for company mentions

Deploying chatbots for automating initial customer claims interviews

Utilizing image recognition for detecting cybersecurity threats

Using text-to-speech tools for marketing material production

14. During hill climbing optimization, a machine learning model struggles with parameter adjustments due to the interdependency of multiple parameters in a complex risk model.

What is the primary drawback of hill climbing in this case?

It can only optimize differentiable functions.

It may miss optimal combinations by adjusting one parameter at a time.

It requires large initial parameter guesses.

It always converges to the global optimum.

15. An insurance company is using a neural network for classifying claims as "Fraudulent" or "Non- Fraudulent." They decide to use ReLU (Rectified Linear Unit) as the activation function for the hidden layers.

What is the primary purpose of using an activation function like ReLU in this neural network?

To scale the output between 0 and 1

To introduce non-linearity

To calculate weights in the hidden layers

To set bias terms to zero

16. A tech company's CXO team is debating how to use AI for internal employee evaluations.

A virtue ethics approach would lead the team to focus on:

Using AI to identify and eliminate the least productive employees

Developing AI systems that encourage and support virtues like honesty, growth, and accountability in employees

Ensuring the AI is cost-effective and legally compliant

Using AI to closely monitor employees to prevent rule violations

17. An analyst is optimizing a linear regression model for predicting salaries based on experience using gradient descent. They start with an initial slope (b1) of 0.8 and use a learning rate of 0.005. In the first iteration, the gradient is calculated as -48.66.

What will be the new value of b1 after this first iteration?

0.8

1.043

0.557

1.059

18. A tech company’s ethics committee is reviewing a hiring algorithm. To ensure procedural fairness, what should the committee prioritize?

Ensuring only the most qualified applicants are hired

Establishing transparent and impartial criteria for all decisions made by the algorithm

Hiring applicants based on recommendations from management

Implementing diversity quotas across departments

19. An investment firm seeks to classify new stock data based on limited labeled examples of previous stocks as "high volatility" or "low volatility." They plan to use this model for classifying future stock data as well.

Which technique should they consider?

Transductive learning with clustering

Inductive learning with self-training

Transductive learning with label propagation

Clustering with k-means

20. A financial services company wants to use a chatbot that can engage in interactive conversations, asking follow-up questions based on customer responses.

Which chatbot type best suits this requirement?

FAQ-based chatbot

Flow-based chatbot

FAQ & flow combination chatbot

Rule-based chatbot

Page 3 of 10

21. A bank wants to minimize the impact of multicollinearity in its linear regression model predicting loan defaults. The data shows high correlation among several predictors.

Which regression technique should they apply to mitigate this issue?

Ordinary Least Squares (OLS) regression

Ridge regression

Logistic regression

Polynomial regression

22. A bank analyst wants to classify a new feedback document using Naïve Bayes. The document contains words previously marked as negative.

Which step should the analyst prioritize to classify this document?

Calculate the posterior probability for each class based on word occurrences.

Remove commonly used words that appear in every document.

Set equal probability for all classes before calculating.

Use only the prior probability of each class to classify.

23. A board director expresses concerns over socio-economic inequality exacerbated by AI adoption.

Which strategy could help address this issue?

Limiting AI to non-revenue-generating functions

Introducing policies to redistribute AI’s benefits through employee profit-sharing

Avoiding the adoption of AI altogether

Allocating all AI profits to research and development

24. A data scientist wants to perform stepwise regression on a dataset with 30 predictors but only 25 observations.

Which approach would be more suitable?

Backward selection because it considers all predictors first

Forward selection because it starts from no predictors

Backward selection because it can handle more predictors than observations

Either method, as both handle large predictor sets equally

25. Which data visualization tool is best suited for assessing whether a dataset follows a normal distribution?

Histogram

Boxplot

Normal Q-Q Plot

Line Chart

26. A bank is considering implementing a new quantitative model for a recently launched product line.

What could be the primary motivation for developing this model?

To maintain alignment with traditional products.

To satisfy existing regulatory requirements without change.

To enhance existing data quality measures.

To respond to business growth and product innovation needs for new model applications.

27. A risk team is developing an NLP tool to monitor news articles for phrases like “credit risk” and “market volatility.”

Why might the team choose to use bi-grams in their analysis?

To capture the individual importance of each word

To handle colloquial expressions in news

To account for phrases with specific meanings when words are together

To remove common words that do not convey meaning

28. A financial institution’s modeling team is evaluating potential future scenarios for a new credit risk model.

What should the team prioritize to improve the model's resilience to market changes?

Rely solely on normal distributions to model all scenarios.

Focus exclusively on recent data trends.

Balance the robustness of the model with adjustments for potential limitations.

Avoid qualitative adjustments for simplicity.

29. A financial institution is implementing a neural network to predict loan default risk. They use gradient descent with backpropagation.

What is the primary purpose of applying the chain rule in this process?

To calculate the initial weights for each neuron.

To calculate the partial derivatives needed to update weights in each layer.

To ensure the model does not overfit the data.

To validate the accuracy of the final model predictions.

30. An insurance firm uses a predictive model to detect fraud. In the interest of balancing sensitivity and specificity, which of the following challenges might the firm face?

Maximizing sensitivity without increasing the false-positive rate

Reducing the base rate to ensure fair outcomes

Ensuring predictive rate parity and equal opportunity simultaneously

Applying demographic parity to improve prediction quality

Page 4 of 10

31. An analyst is using a cloud-based LLM to generate summaries of financial documents. The analyst wishes to use a model that adapts to new tasks with minimal examples.

Which of the following describes this adaptation method?

Low-shot task transfer

Batch processing

Hyperparameter tuning

Parameter pruning

32. In a regression model y=β0+β1x+u, if the parameter β0 (intercept) is estimated to be 2.5, what is the interpretation of this estimate?

The change in y for each unit change in x.

The expected value of y when x is zero.

The average error in the model.

The constant variance of the error term u.

33. A risk model has been operational for several months.

To evaluate its ongoing effectiveness in supporting decision-making, what approach should be taken?

Conduct unit testing on individual components.

Focus only on statistical measures of model accuracy.

Perform a use test to assess if the model is accepted, trusted, and applied effectively by decision- makers.

Review historical data for outliers.

34. A multinational healthcare provider is using AI to prioritize patient access to specialist appointments.

How should the provider apply justice in this scenario?

Prioritize patients based on their insurance status

Allocate appointments based on medical need and urgency, ensuring fair access to all patients

Offer priority access to patients who have been with the clinic for longer

Use an algorithm that prioritizes patients from wealthier areas for quicker access

35. A financial analyst wants to apply a regularized regression model that reduces extreme coefficient values without eliminating any features, as they all provide valuable insights.

Which regularization method is most appropriate?

LASSO

Ridge

Elastic Net

Stepwise selection

36. A Board of Directors at a global tech firm is considering developing a data-collection AI that could enhance user experience but involves intrusive tracking.

According to rule consequentialism, how should they decide?

Approve the AI if it increases company revenue and shareholder value

Approve the AI if it adheres to rules that generally maximize trust and benefit users

Reject the AI if it does not comply with all regional data privacy laws

Reject the AI as it could potentially reduce the number of future data projects

37. An image recognition startup is using a neural network with ReLU activation in the hidden layers. However, they are noticing that some neurons become inactive and only output zero values regardless of the input.

What activation function modification could help prevent this issue?

Sigmoid

Softmax

Leaky ReLU

Hyperbolic tangent

38. In cost complexity pruning, a bank uses a parameter, , to balance the residual sum of squares (RSS) and the number of terminal nodes.

What is the primary role of this parameter?

Maximize the training accuracy

Minimize the test set error

Control the trade-off between tree size and fit

Increase the complexity of the tree

39. An investment firm is building a machine learning model that uses alternative data to make market predictions.

To avoid the risk of model destruction, what should the firm prioritize?

Confirming the legal right to use each data source included in the model.

Optimizing the model to maximize predictive accuracy.

Using only data that is readily available online without additional verification.

Ensuring the data is recent to improve model reliability.

40. A financial analyst is evaluating dimensionality reduction methods for a Treasury yield dataset using both PCA and autoencoders. The analyst wants to keep the number of components low but still capture essential data structure.

Which approach might allow for more flexibility in capturing non-linear relationships?

PCA with more components

PCA with fewer components

Autoencoder with fewer hidden units

Autoencoder with more hidden units

Page 5 of 10

41. A fintech company wants to classify images of check deposits into fraudulent and non-fraudulent categories.

Given that each image has high pixel density, which neural network architecture would be most suitable for this task?

Recurrent Neural Network (RNN)

Feedforward Neural Network

Convolutional Neural Network (CNN)

Long Short-Term Memory (LSTM)

42. A financial firm is using SVM with multiple features to classify potential investments as "High Return" or "Low Return." They notice that not all data points fit a clear boundary and some overlap exists.

What can the firm do to improve the model's performance on this non-linearly separable data?

Use a linear SVM model with strict margin constraints

Apply a non-linear kernel to the SVM model

Decrease the number of features to reduce complexity

Only use the data points that fit within a clear boundary

43. A data scientist is working with financial transaction data where transaction amounts vary significantly, ranging from a few cents to thousands of dollars. She decides to apply feature scaling to improve model performance.

Which approach should she use if her data contains several extreme outliers?

Apply normalization to scale all values between 0 and 1.

Apply standardization to ensure the data has a mean of zero and unit variance.

Remove all outliers before scaling.

Leave the data as-is, as scaling will not help.

44. A data science team is exploring different cloud-based LLMs for text classification without additional task-specific examples.

Which model generation first enabled “no-shot fine-tuning”?

GPT-1

GPT-2

GPT-3

Chat-GPT

45. To manage AI-related risks, a company decides to develop an ethical AI framework.

Which approach would enhance the effectiveness of this framework?

Developing the framework internally without external input

Consulting with AI experts, ethicists, and stakeholders

Focusing only on high-level principles without specific guidelines

Prioritizing company profit motives over ethical considerations

46. A financial institution is developing a data protection strategy to comply with regulatory requirements.

Which action should be prioritized to ensure data security and integrity?

Allow all employees unrestricted access to data for ease of workflow.

Focus primarily on external threats, assuming internal data access is secure.

Retain all data indefinitely to avoid losing potentially valuable information.

Implement a comprehensive data protection plan that includes access controls, a data retention policy, and measures to prevent data corruption or loss.

47. A data scientist must choose between a highly interpretable decision tree and a less interpretable ensemble model for predicting loan defaults. The ensemble model has higher prediction accuracy.

Which model is more suitable if the goal is explaining to stakeholders why certain loans were flagged as high risk?

The ensemble model, as it has higher prediction accuracy

The decision tree, as it is easier to interpret

Both models are equally suitable

Neither model is suitable for explainability

48. In the setup of a QRM, which step is responsible for integrating the model core into an existing system, allowing it to collect data, process outputs, and schedule computations?

System implementation

Model design

Core implementation

Interface testing

49. A credit risk analyst is evaluating a dataset of 200 borrowers, of whom only 20 defaulted on their loans. To perform model validation, they want to ensure that each cross-validation fold contains an equal proportion of defaulted and non-defaulted borrowers.

Which method should they choose?

Standard k-fold cross-validation

Stratified k-fold cross-validation

Leave-One-Out Cross-Validation (LOOCV)

Bootstrapping

50. A model developer is tasked with creating a model for evaluating integrals in financial risk modeling.

What is an essential step they should take when implementing numerical evaluation to maintain reliability?

Assume that the model’s theoretical basis ensures reliability without testing.

Maximize computational efficiency by reducing step size to a minimum.

Ignore convergence issues if they appear minor.

Perform extensive testing to confirm convergence and stability in practical applications.

Page 6 of 10

51. A financial institution is testing a new QRM for its trading platform.

Which type of testing is crucial to confirm that all integrated components of the model work correctly together?

Unit testing

Component testing

Integration testing

Regression testing

52. A credit analyst is creating word vectors from credit reviews to classify them by risk level. They find that vectors from high-risk reviews are dominated by repeated terms like “default” and “risk.”

Which method can help prevent this overrepresentation from skewing the results?

Adjust the vocabulary to exclude common terms

Apply L1 normalization to word vectors

Use L2 normalization to scale the vectors

Exclude words with frequencies above a certain threshold

53. A credit card company wants to reduce the dimensionality of customer transaction data to identify spending patterns without using predefined labels.

Which technique would be most suitable?

K-means Clustering

Principal Component Analysis (PCA)

Autoencoder

Support Vector Machine

54. A bank is using a neural network with a single hidden layer to predict loan default probability. The team notices the network is performing similarly to a linear regression model and not capturing complex patterns.

What might be the cause?

Too many features in the input layer

Absence of a bias term

Lack of non-linear activation function

Too many neurons in the hidden layer

55. A technology company in China is handling user data under the Personal Information Protection Law (PIPL).

What primary principle should the company follow?

Maximize data collection for comprehensive user profiles.

Share data freely with affiliated companies to improve user experience.

Minimize data collection, obtain user consent, and ensure data transparency.

Store data indefinitely to anticipate future business needs.

56. An admissions committee decides to incorporate a measure of resilience alongside grades in its AI- based selection criteria.

Which challenge is the committee likely to face with this approach to individual fairness?

Defining resilience as a measurable trait

Avoiding direct discrimination

Ensuring grades remain the sole basis for admissions

Guaranteeing equal admissions for all applicants

57. A retail company uses an algorithm to set product prices, which unintentionally raises prices in neighborhoods with a high concentration of minority groups.

Which factor determines if this is considered discrimination?

If the algorithm's outcome affects revenue

If the price difference lacks a valid business justification

If the company makes the algorithm publicly accessible

If customers in those neighborhoods are unaware of the pricing model

58. A financial analyst uses a model to predict monthly returns on an investment portfolio and wants to evaluate the model’s performance.

Which of the following metrics is most appropriate if the analyst wants to avoid overemphasizing large deviations?

MSE

RMSE

MAE

MAPE

59. An insurance company uses two different AI models for policy approval, one for Group A and another for Group B.

The confusion matrices for each group are as follows:

Group A Confusion Matrix:

• True Positives: 50

• False Positives: 10

• False Negatives: 40

• True Negatives: 90

Group B Confusion Matrix:

• True Positives: 30

• False Positives: 20

• False Negatives: 70

• True Negatives: 80

What fairness technique should the company use to balance the performance between the two groups?

Equalize the number of true positives for both groups.

Adjust thresholds to achieve similar false positive rates for both groups.

Apply a higher penalty for false negatives in Group

Implement demographic parity by ensuring equal acceptance rates for both groups.

60. Following a backlash over an AI’s biased hiring decisions, which step could best help a company restore public trust?

Release an apology and halt all AI hiring processes

Implement more diverse and representative data sources and disclose improvements

Ignore the incident and focus on internal improvements

Use third-party AI providers to avoid direct involvement

Page 7 of 10

61. A government regulator is reviewing an AI system for public health data analysis.

To align with nonmaleficence, what should the regulator ensure before approval?

That potential risks are justified by significant public health benefits

That any risks are only minor and are not fully disclosed to the public

That the system is implemented as quickly as possible to show efficiency

That the public provides consent without being informed of any risks

62. A political organization wants to use AI to target potential voters with tailored ads.

Which approach raises the least concern about manipulation?

Designing ads based on users’ personality profiles without their knowledge

Providing users with clear information about the nature of the targeted ads

Using historical voting behavior to target ads without disclosing data sources

Collecting data through personality quizzes to personalize ads without transparency

63. The Board is concerned about talent retention, as recent surveys show employees are increasingly concerned with the company’s ethics.

How could implementing a practical ethics framework benefit talent retention?

It would have minimal impact on employee satisfaction and retention

It would attract talent that values a company’s commitment to ethical practices

It would allow the company to prioritize efficiency over employee concerns

It would reduce transparency in employee decision-making

64. After public outcry over a biased AI model, which of the following steps should a company prioritize to rebuild trust?

Discontinue use of AI in all applications

Communicate changes in data sources and fairness improvements

Blame the software vendor and defer responsibility

Avoid using data from underrepresented groups

65. A healthcare provider uses a third-party AI system to assist in diagnosis but cannot access the algorithm's inner workings due to confidentiality agreements.

What risk does this pose?

Users will have full control over the AI system

The provider can easily modify the algorithm

Lack of transparency may prevent understanding of bias or error risks

The provider can independently train the AI system on local data

66. A bank's Board of Directors is discussing the implementation of an AI tool for credit scoring that uses

detailed personal data. According to deontological ethics, what should be the main consideration?

The tool should not be used if it violates customer privacy rights, even if it is more accurate

The tool should be used if it improves financial outcomes for the bank

The tool should be implemented as long as it is legal, regardless of privacy concerns

The tool should prioritize cost savings over customer data privacy

67. A financial firm is conducting white-box testing on a newly developed risk model.

What does this type of testing involve?

Reviewing the internal code structure and algorithms of the model.

Running test cases without knowledge of the model’s internal structure.

Using only historical data for testing.

Ignoring code details and focusing on user interface testing.

68. Regulators are concerned with the rapid development of AI and its associated risks.

What is a key challenge they face in managing these risks?

Ensuring that all AI applications are profitable

Implementing governance mechanisms that keep pace with AI developments

Guaranteeing AI models are transparent to all users

Preventing AI from being used in customer service roles

69. A tech firm is evaluating different models for an NLP project requiring high computational efficiency and parallel processing.

Why might transformers be more suitable than RNNs for this task?

They use contextual embeddings

They are sequence-dependent

They allow for parallel processing

They rely on sequence ordering

70. The Chief Risk Officer (CRO) is reviewing potential risks in a new AI-based product.

How could a practical ethics framework assist in managing these risks?

It allows the company to ignore ethical concerns in favor of regulatory compliance

It encourages the integration of ethical principles, which can help identify and mitigate safety and compliance risks

It limits the company’s innovation by restricting certain technologies

It focuses only on avoiding legal consequences

Page 8 of 10

71. A retail bank uses self-training to classify loan applicants as high or low risk, but it finds that updating the model after each new labeled data point is computationally intensive.

Which approach can the bank use to reduce this burden?

Select multiple best-predicted instances in each round

Use an unsupervised learning model instead

Increase the complexity of the initial model

Use transductive learning instead

72. An analyst applies the K-means algorithm to cluster financial transactions and chooses Euclidean distance as the measure for assigning points to centroids.

If the data has features with different scales (e.g., transaction amount in thousands and frequency as an integer), what should the analyst do before running K-means?

Nothing, since K-means is unaffected by feature scales

Use Manhattan distance instead of Euclidean distance

Normalize or standardize the data

Increase the number of clusters (K)

73. A health services provider is reviewing its data management practices. To protect patient data ethically, which step should it take?

Store data in easily accessible files for operational efficiency.

Retain patient data indefinitely in case it’s needed for future treatments.

Encrypt sensitive data and restrict access to only essential staff members.

Allow data to be accessed and transferred by external partners for cost-sharing purposes.

74. A justice system uses an AI model to predict failure-to-appear (FTA) rates in court.

To ensure fairness without explicitly using protected group data, what could be a potential unintended outcome?

The model underestimates FTA for all demographics

The model’s accuracy decreases for high-risk groups

The model’s reliance on correlated variables increases potential bias in predictions

The model achieves demographic parity but lacks predictive accuracy

75. In a deep reinforcement learning model for portfolio optimization, the agent uses neural networks to estimate Q-values for different actions in various market states.

What primary purpose does the neural network serve in this scenario?

To generate possible future actions for a given state

To approximate the Q-value function, reducing the need for a large lookup table

To improve the agent's policy using a tabular approach

To prioritize certain actions based on historical performance

76. A real estate firm uses a regression tree to predict house prices based on features such as house age and proximity to a metro station.

Which of the following objectives is the primary focus when building this regression tree?

Minimizing the number of splits in the tree

Maximizing the average price prediction in each region

Minimizing the residual sum of squares (RSS) within each region

Maximizing the variance between regions

77. A financial analyst is using logistic regression to predict the probability of a loan default based on borrower characteristics. The output probability ranges between 0 and 1.

What transformation allows this model to constrain the output in this range?

Linear transformation

Logistic transformation

Exponential transformation

Polynomial transformation

78. An analyst is examining customer satisfaction levels (low, medium, high) to develop a predictive model.

Which type of model should the analyst use to incorporate the ordered nature of satisfaction levels?

Multinomial logit model

Ordered logit model

Binary logit model

Linear regression model

79. In assessing whether an AI hiring tool adheres to “fairness through unawareness,” which of the following scenarios suggests that this approach may fail to prevent bias?

The AI model uses all available data except race and gender

The AI model applies a fairness constraint to ensure equal opportunity

The AI model’s predictions weigh correlated variables more heavily in absence of protected data

The AI model only considers qualifications directly relevant to job performance

80. A model estimates the relationship between interest rates and loan demand using simple linear regression. The model’s prediction equation is y=β0-β1x+u. If β1 = 2, what does this imply about the relationship between interest rates and loan demand?

For each unit increase in interest rate, loan demand increases by 2 units.

For each unit decrease in interest rate, loan demand decreases by 2 units.

For each unit increase in interest rate, loan demand decreases by 2 units.

β1 cannot be negative in a linear model.

Page 9 of 10

81. A bank is implementing a quantitative risk model (QRM) and needs to ensure that the model is designed in a way that can be accurately translated into code.

Which implementation task should the bank prioritize first?

Model design

Core implementation

System implementation

Post-implementation review

82. The CEO is concerned about the societal impact of a new AI-driven social media algorithm that recommends content. From a practical ethics viewpoint, what approach should the company take?

Design the algorithm to prioritize user engagement above all

Incorporate stakeholder feedback to align the algorithm with societal values

Allow the algorithm to operate without restrictions to maximize revenue

Use only deontological principles in its design for strict ethical compliance

83. An e-commerce company is deciding on a data retention policy for customer purchase data.

What policy would best address ethical concerns related to privacy?

Store all customer data permanently for analysis and future use.

Regularly delete customer data that is no longer needed for business purposes.

Retain data only if customers explicitly ask for it to be saved.

Use minimal encryption to save storage costs.

84. An investment firm has observed an increase in the portfolio size associated with a low-risk model.

What is an appropriate action to manage model risk?

Reassess the model’s risk ranking and impose limits if needed.

Ignore the change as long as the model remains low-risk.

Apply the model’s original assumptions regardless of portfolio growth.

Increase the model’s usage scope without further validation.

85. A bank is using a multinomial logit model to predict customers' choice among different loan products: personal loan, home loan, or credit card.

If personal loan is set as the baseline category, which statement correctly describes the probability of choosing a credit card?

The probability is the same as the probability of not choosing a home loan

It must be calculated as a separate probability from the baseline

It is one minus the probability of choosing a personal loan

It is calculated as the product of probabilities for each category

86. A data scientist is working on a credit scoring model for a bank. One of the variables they want to include is the applicant's region of residence in the US, which has five categories: Pacific, Rocky Mountain, Midwest, Northeast, and South.

Which of the following approaches is most appropriate for encoding this categorical variable to avoid issues with the model?

Assign a single variable with values such as Pacific = 0, Rocky Mountain = 1, Midwest = 2, Northeast = 3, and South = 4.

Create a dummy variable for each region (Pacific, Rocky Mountain, Midwest, Northeast, South), with each applicant having a value of 1 for their region and 0 for others.

Create a single variable with binary encoding, assigning each region a unique binary code (e.g., Pacific = 000, Rocky Mountain = 001, Midwest = 010, etc.).

Use one-hot encoding for the regions but exclude one category (e.g., exclude "Pacific") to avoid multicollinearity in the regression model.

87. A bank is using a nonlinear model for credit scoring, and the model’s parameters are optimized via NLS. After each iteration, the residual sum of squares (RSS) decreases, but the decrease is very small.

What does this indicate about the model’s optimization process?

The model has reached an optimal solution.

The initial parameter estimates were incorrect.

The learning rate in the optimization is likely too high.

The model is likely at or near a local minimum.

88. A bank is developing an AI-based credit risk model.

What is a key step in the data collection and preprocessing stage to ensure robust model development?

Allow the model to exclude data automatically without human intervention.

Ignore missing values to expedite the data preparation process.

Use only recent data to improve model performance.

Document any data exclusions and review excluded data for potential insights before proceeding.

89. A bank wants to analyze long text sequences to understand customer sentiment over time. However, their RNN model faces difficulty in learning from words far apart in sentences.

What model modification could help with long-range dependencies?

Increase the training epochs

Switch to a Convolutional Neural Network

Use a Long Short-Term Memory (LSTM) network

Reduce the sequence length

90. A research team is training an LLM for sentiment analysis. They need the model to avoid bias and accurately represent positive and negative sentiment data.

Which step in the LLM training process is essential for achieving this?

Tokenization

Data balancing

Positional encoding

Masking

Page 10 of 10

91. A QRM development team has created a market risk model for derivative products.

Which approach should they use to prevent underestimating risks?

Use advanced quantitative methods exclusively.

Focus only on quantitative risk measures.

Rely solely on correlations without additional risk analysis.

Utilize both qualitative and quantitative methods.

92. A model developer is implementing Taylor series approximations to simplify market risk calculations for a portfolio with small changes in risk factors.

What should they prioritize to avoid misuse of this approximation?

Apply the Taylor series approximation for all financial instruments.

Assume that the approximation will work for any risk factor size.

Limit the use of Taylor series to simple instruments and minor changes in risk factors.

Use the approximation without further validation steps.

93. In designing an AI for a two-player game, which strategy allows the AI to choose the move that minimizes the maximum advantage an opponent can gain?

Alpha-beta pruning, which restricts the number of moves considered by the A

Reinforcement learning, which uses rewards to guide future moves.

Minimax search, where the AI assumes the opponent will always choose the best possible response.

Monte Carlo simulation, which tests random move sequences to determine probabilities of winning.

94. A company plans to implement a third-party recruitment algorithm, yet they realize the algorithm’s definition of “good employee” may differ from their own.

What should the company do to ensure alignment with its values?

Apply the algorithm in a limited capacity first

Regularly audit the algorithm’s outputs for fairness

Customize the algorithm to reflect the company’s values

Use the algorithm as is to avoid custom development costs

95. A healthcare provider faces an accuracy-interpretability trade-off when choosing between a complex AI model and a simpler, interpretable model for diagnosis.

Which of the following would be the most practical reason for choosing the complex model over the interpretable one?

The complex model can explain itself to any user

The complex model achieves significantly higher diagnostic accuracy

The complex model is easier to audit

The complex model requires less computational power

96. A company faces criticism for a privacy breach involving customer data used in an AI model.

What corrective action can help regain customer trust?

Increase data collection to improve model accuracy

Strengthen data privacy measures and notify customers about improvements

Make data collection anonymous without further action

Transfer data storage to a third-party provider

97. Why was the “frame problem” a significant challenge for classical AI?

It involved coding all possible actions for every scenario.

It required tracking changes and stable aspects within a situation after an action was taken.

It prevented AI from processing perfect information games like chess.

It made AI slower in games by increasing the number of moves it considered.

98. A bank’s risk management team is using discretization in a model with discrete time intervals.

How can the team ensure that the discretization does not introduce significant errors?

Set parameters such as time step/bucket size to control error.

Focus only on reducing the running time of the model.

Avoid parameter adjustments for simplicity.

Use closed-form solutions only.

99. A portfolio manager is using reinforcement learning to maximize returns over time.

To evaluate each policy, he wants to calculate the long-term value of returns using the Bellman equation, considering both immediate rewards and discounted future rewards.

Which function does he use to achieve this?

Policy function

Reward function

Value function V(s)

State-action function Q(s,a)

100. A public health organization uses an algorithm trained on pre-COVID data to identify at-risk patients. Despite this, the model's predictions are found to be increasingly inaccurate.

Which of the following is likely causing the issue?

The model’s architecture is outdated

The data used to train the model is no longer representative

The algorithm has a high degree of interpretability

The algorithm is designed for a different healthcare domain

TAGS:

RAI, RAI exam dumps

Notify of

Label

Name*

Email*

Website

Label

Name*

Email*

Website

0 Comments

Oldest

Newest Most Voted

Inline Feedbacks

View all comments

Posts

Test Online 2016-FRR Dumps Questions to Be Certified

To help prepare for the 2016-FRR exam, you may want to consider using 2016-FRR exam dumps questions. These questions are...

Get RAI Dumps Full Version

Q&As: 330
Versions: PDF and Software Download Now

About Dumpsinfo

Dumpsinfo is a good platform providing the latest exam information and dumps questions for all IT certification exams. You can study all the latest exam dumps questions online.

[email protected]

Mon - Sat 9:00am - 6:00pm