GARP RAI: History and Overview of AI Concepts

Use this page to isolate History and Overview of AI Concepts before returning to mixed GARP RAI practice.

Topic snapshot

Sample questions

These questions are original Finance Prep practice items aligned to this topic area. They are designed for self-assessment and are not official exam questions.

Question 1

Topic: History and Overview of AI Concepts

A bank uses an AI model to rank small-business loan applications for manual underwriter review. The model was validated before launch, but the applicant mix has changed after a new digital channel, and management wants to continue using the model without changing its intended use. Which evidence would best support whether the model remains fit for that intended use?

• A. The original validation report showing strong test-set performance before the digital channel was introduced
• B. Recent monitoring results on current production applications comparing predicted risk rankings with realized outcomes against approved performance criteria
• C. Confirmation that the model code and hyperparameters have not changed since approval
• D. A business report showing that underwriters find the model’s risk rankings convenient to use

Best answer: B

What this tests: History and Overview of AI Concepts

Explanation: A model remains fit for intended use when evidence shows it still performs adequately for the current population, decision context, and approved performance expectations. Because the applicant mix has changed, historical validation alone is insufficient. The best evidence is recent monitoring or back-testing on current production data, comparing predictions with realized outcomes and approved criteria such as discrimination, calibration, error rates, or ranking effectiveness. Unchanged code does not prove unchanged performance, because data drift or population shift can degrade results. User convenience may support adoption, but it does not establish model performance or risk fitness.

• Original validation is useful baseline evidence, but it may no longer reflect the changed applicant population.
• Unchanged code misses the risk that inputs or borrower behavior have shifted.
• Underwriter convenience addresses usability, not whether predictions remain accurate or reliable.

Current, outcome-based performance evidence tied to the model’s intended use is the strongest basis for assessing ongoing fitness.

Question 2

Topic: History and Overview of AI Concepts

A bank uses an AI score to support credit line-increase approvals. Although the model has strong overall accuracy, the risk team evaluates false approvals, false declines, and cutoff choices in terms of expected losses, revenue, and customer impact for that specific approval decision. Which concept does this description best illustrate?

• A. Feature engineering for predictive signal
• B. Unsupervised pattern discovery
• C. Decision-centered performance evaluation
• D. Model transparency through explainability

Best answer: C

What this tests: History and Overview of AI Concepts

Explanation: Model performance should be judged in relation to the decision the model informs, not only by generic statistical metrics. A credit approval model may have high overall accuracy but still be unsuitable if its errors create unacceptable credit losses, missed revenue, customer harm, or control issues. Decision-centered evaluation connects metrics such as false positives, false negatives, thresholds, and calibration to the business action being taken. This helps determine whether the model is fit for purpose in its operating context.

• Unsupervised pattern discovery concerns finding structure in unlabeled data, not evaluating decision consequences.
• Feature engineering focuses on creating useful input variables, not judging whether outcomes support a business decision.
• Explainability helps users understand model outputs, but it is not the same as evaluating performance against decision-specific costs and benefits.

It assesses model performance by the consequences and trade-offs of the business decision the model supports.

Question 3

Topic: History and Overview of AI Concepts

A bank trains an AI classifier to predict whether loan applicants will default. The training data includes only applicants who were previously approved, because rejected applicants have no repayment outcomes. The model will now be used to classify all new applicants, including profiles that were often rejected in the past. Which data-quality issue most directly undermines the classification outcome?

• A. Label noise
• B. Target leakage
• C. Sample selection bias
• D. Class imbalance

Best answer: C

What this tests: History and Overview of AI Concepts

Explanation: Sample selection bias occurs when the observations used for training are not representative of the population where the model will be applied. Here, the classifier learns only from previously approved borrowers, while the intended scoring population includes all new applicants. Because rejected applicants are systematically missing, the model may learn relationships that do not hold for applicant types it has not observed with repayment outcomes. This can undermine default predictions even if the historical approved-applicant records are accurate. The key issue is not simply model choice; it is a data coverage and representativeness problem tied to how the training sample was created.

• Target leakage involves using information that would not be available at prediction time; the stem instead describes missing segments of the target population.
• Class imbalance concerns one outcome class being much rarer than another; no default-rate imbalance is stated.
• Label noise means labels are incorrect or unreliable; the issue here is that labels are unavailable for excluded rejected applicants.

The training data excludes rejected applicants, so it may not represent the full population the classifier will score.

Question 4

Topic: History and Overview of AI Concepts

A consumer-lending risk team is asked to identify accounts most likely to become 60 days past due in the next 90 days. The team currently has a dashboard that summarizes last quarter’s delinquencies by product and region. Management wants to prioritize early outreach before borrowers miss additional payments. What is the BEST action for the team?

• A. Build and validate a model that uses historical account attributes and later delinquency outcomes to score current accounts for next-90-day risk.
• B. Prepare a narrative report explaining the operational causes of last quarter’s delinquency increase.
• C. Expand the dashboard to show last quarter’s delinquency rates by branch, product, region, and customer segment.
• D. Rank accounts by their current number of missed payments without testing whether the ranking predicts 90-day delinquency.

Best answer: A

What this tests: History and Overview of AI Concepts

Explanation: Descriptive analytics summarizes or explains what has already happened, such as delinquency rates by product, region, or branch. That can support management understanding but does not directly forecast which current accounts are likely to become delinquent. Because the objective is to prioritize outreach before future missed payments, the better fit is a predictive model. The model should use historical borrower and account features, connect them to subsequent delinquency outcomes, and be validated on its ability to predict future risk. A simple ranking rule may be useful as a baseline, but without validation against future outcomes it is not sufficient for a forecasting objective.

• Expanding the dashboard improves descriptive detail, but it still focuses on past delinquency patterns.
• Explaining last quarter’s causes supports retrospective analysis, not account-level forecasting.
• Ranking by missed payments may be intuitive, but it is not a validated predictive approach unless tested against future outcomes.

Forecasting future delinquency requires a predictive model trained and tested against future outcome labels, not only summaries of past activity.

Question 5

Topic: History and Overview of AI Concepts

A bank plans to use a supervised model to prioritize credit-card accounts for early outreach. The model has 91% overall accuracy on a holdout sample, but outreach staff can contact only 10% of accounts each week, and missing a truly at-risk account is much more costly than contacting a low-risk account. What is the best action before approving the model for this use?

• A. Evaluate performance at the proposed outreach threshold using business-relevant measures such as top-decile recall, lift, and expected cost-benefit.
• B. Require retraining until overall accuracy reaches a higher target, then approve the model if that target is met.
• C. Limit the review to confirming that the model inputs are available before each outreach cycle.
• D. Approve the model because high overall holdout accuracy demonstrates adequate predictive performance.

Best answer: A

What this tests: History and Overview of AI Concepts

Explanation: Model performance should be evaluated in relation to the business decision the model supports. In this case, the model is not being used simply to classify all accounts correctly; it is being used to rank accounts for a limited outreach capacity. Overall accuracy may hide poor performance on the most important cases, especially when the cost of missed at-risk accounts is high. The appropriate review links technical metrics to the decision context: which accounts are selected, how many can be contacted, what errors matter most, and whether the selected threshold improves business outcomes compared with alternatives.

• High overall accuracy is tempting, but it may not show whether the model identifies the riskiest 10% of accounts.
• Raising an arbitrary accuracy target does not address the asymmetric cost of errors or operational capacity.
• Input availability is necessary for implementation, but it does not establish that the model improves the outreach decision.

The model should be judged by how well it supports the actual prioritization decision under capacity and cost constraints.

Question 6

Topic: History and Overview of AI Concepts

A bank’s analytics team has a large set of customer transaction records, but the records do not contain preassigned outcomes such as fraud or not fraud. The team wants the model to identify natural groupings and unusual patterns in the data. Which machine learning concept best matches this training approach?

• A. Unsupervised learning
• B. Rule-based expert system
• C. Supervised learning
• D. Reinforcement learning

Best answer: A

What this tests: History and Overview of AI Concepts

Explanation: The decisive fact is that the training data do not include known target labels such as fraud or not fraud. When a model learns from unlabeled data to find structure, clusters, or anomalies, the approach is unsupervised learning. By contrast, supervised learning requires labeled examples so the model can learn a mapping from input features to a known outcome. In this scenario, the bank is not training the model to reproduce known fraud decisions; it is asking the model to discover patterns in the transaction data.

• Supervised learning would fit if each historical transaction included a known outcome label used as the training target.
• Reinforcement learning involves learning actions through rewards or penalties, not simply grouping unlabeled records.
• A rule-based expert system applies predefined human-written rules rather than learning patterns from data.

Unsupervised learning uses unlabeled data to discover patterns, clusters, or anomalies without target outcomes provided during training.

Question 7

Topic: History and Overview of AI Concepts

A bank’s financial-crime operations team wants to reduce manual review time for transaction-monitoring alerts. It has several years of historical alerts with final dispositions labeled as true suspicious activity or false positive, and it wants new alerts routed to review queues based on the likelihood of being truly suspicious. Which model purpose is the best fit?

• A. Use a generative language model to draft alert narratives before determining alert disposition.
• B. Use a supervised classification model to estimate each alert’s likelihood of being a true suspicious-activity case.
• C. Use a regression model to forecast the monthly number of alerts across the monitoring unit.
• D. Use an unsupervised clustering model to find customer peer groups without using the alert labels.

Best answer: B

What this tests: History and Overview of AI Concepts

Explanation: The use case asks the model to assign or score new alerts against known labeled outcomes: true suspicious activity versus false positive. That is a supervised classification purpose because the model learns from examples where the correct class is already known and applies that pattern to new cases. The result can support queue routing, prioritization, and reviewer efficiency, while human review and controls remain important in a compliance setting. The other model purposes address different tasks: forecasting quantities, discovering unlabeled groups, or generating text. Those may support related workflows, but they do not directly solve the stated objective of predicting the disposition category for each new alert.

• Forecasting monthly alert volume is a planning use case, not a per-alert disposition decision.
• Clustering can reveal segments or peer groups, but it ignores the available true/false labels needed for this task.
• Generating narratives may assist documentation, but it does not determine whether an alert is likely suspicious.

The target outcome is a labeled category, so supervised classification best matches the routing objective.

Question 8

Topic: History and Overview of AI Concepts

A bank deploys an AI model that compares new payment activity with historical customer patterns and highlights transactions that look unusual for analyst review. The model is intended to flag deviations from expected behavior, not to automatically approve, rank, or forecast transactions. Which AI concept best matches this purpose?

• A. Classification model
• B. Regression model
• C. Recommender model
• D. Anomaly-detection model

Best answer: D

What this tests: History and Overview of AI Concepts

Explanation: Anomaly-detection models are used to identify observations, events, or patterns that differ materially from what is expected or typical. In financial services, they are often used to flag unusual transactions, account activity, system behavior, or operational events for review. The key point is that the model raises an alert or exception for investigation; it does not necessarily prove fraud or make a final decision. In the stem, the model compares new payment activity with historical customer behavior and highlights deviations, which directly matches anomaly detection.

• A classification model assigns observations to predefined categories, such as fraud or not fraud, rather than primarily flagging unusual deviations.
• A regression model predicts a numeric value, such as loss amount or transaction volume, rather than identifying outliers.
• A recommender model suggests items or actions based on preferences or similarity, not unusual activity for investigation.

An anomaly-detection model is designed to identify unusual patterns or outliers for further investigation.

Question 9

Topic: History and Overview of AI Concepts

A bank uses a delinquency-workflow tool to route accounts to collections queues. Business analysts wrote explicit if/then rules, such as assigning accounts over 30 days past due to a senior queue; the tool has no training dataset, learned parameters, or periodic retraining, though it logs outcomes for reporting. During an AI inventory review, the first line proposes labeling it as a machine-learning model because it processes customer data. What is the best action for the risk reviewer?

• A. Require model retraining before deployment because routing accuracy can only be improved through learned parameters.
• B. Classify it as rules-based automation and apply appropriate control oversight, rather than treating it as a machine-learning model.
• C. Exclude it from all governance review because systems without learned logic create no AI-related operational risk.
• D. Classify it as supervised machine learning because the rules use historical delinquency outcomes.

Best answer: B

What this tests: History and Overview of AI Concepts

Explanation: Rules-based automation applies logic that humans explicitly specify, such as if/then decision rules. A machine-learning model, by contrast, learns patterns or parameters from data during training and then uses those learned relationships to make predictions, classifications, or decisions. In this scenario, the tool processes customer data and logs outcomes, but those facts do not make it machine learning. The decisive fact is that the routing logic was manually coded and there is no training process or learned parameter set. The best action is to classify it accurately as rules-based automation while still applying relevant controls, such as change management, testing, access control, and monitoring appropriate to its business impact.

• Processing customer data does not by itself make a system a machine-learning model.
• Historical outcome logs may support reporting or future analysis, but they are not training unless the system learns decision logic from them.
• Retraining is not applicable when routing is driven by fixed coded rules rather than learned parameters.
• Non-ML automation can still create operational, fairness, compliance, or customer-impact risks that require governance.

The decision logic is explicitly coded by humans and is not learned from data, so it is rules-based automation even though it uses customer data.

Question 10

Topic: History and Overview of AI Concepts

A bank uses an AI system trained and validated only to classify incoming payment messages as likely duplicate or not duplicate. It performs well in that workflow but cannot answer customer-service questions, summarize policies, or perform other banking tasks without separate design and training. Which concept best describes this system?

• A. Artificial general intelligence
• B. General-purpose AI
• C. Foundation model
• D. Narrow task-specific AI

Best answer: D

What this tests: History and Overview of AI Concepts

Explanation: Narrow or task-specific AI is designed to perform a defined function within a limited context, such as classifying payment messages for duplicate risk. Strong performance on that task does not imply the system can reason across domains, adapt to unrelated workflows, or perform open-ended business activities. Broader general-purpose AI capabilities involve models that can be adapted across many tasks, and artificial general intelligence would imply human-like flexibility across domains. In this scenario, the need for separate design and training before handling other banking tasks is the key evidence that the system is narrow AI.

• General-purpose AI would imply broader adaptability across multiple tasks, which the system does not have.
• Artificial general intelligence would imply flexible, human-like capability across domains, far beyond the described classifier.
• A foundation model may support many downstream tasks, but the described system is a single-purpose classifier rather than a broad pretrained model.

The system is optimized for one defined business task and lacks broader cross-domain capabilities.

Continue with full practice

Use the GARP RAI Practice Test page for the full Finance Prep practice bank, mixed-topic practice, timed mock exams, and explanations.

Related focused pages

• Free GARP RAI Full-Length Practice Exam
• GARP RAI: AI Tools and Techniques
• GARP RAI: Risks and Risk Factors
• GARP RAI: Responsible and Ethical AI
• GARP RAI: Data and AI Model Governance

Free review resource

Use the full Finance Prep practice page above for the latest review links and practice page.