50 IBM Assessment: Foundations of AI Practice Questions: Question Bank 2025

A retail company wants to add an AI feature that automatically routes customer emails to the correct support team (billing, shipping, returns). Which AI capability best fits this requirement?

In a discussion about AI, a stakeholder says: "Our system should learn patterns from historical data to make predictions without being explicitly programmed with rules." Which concept are they describing?

A team is building an AI model and wants to reduce the risk of overfitting. Which action is a recommended best practice?

A product manager asks why a conversational AI solution needs "training data" for intent recognition. What is the best explanation?

A data scientist reports: "Our model performs extremely well on the training set but poorly on new data." Which situation is most likely occurring?

A healthcare organization wants to use an AI model to assist clinicians, but it must provide understandable reasons for its predictions to support clinical decision-making. Which requirement does this describe?

A team is deploying an AI model into a production application. Which approach best supports continuous monitoring and governance after deployment?

A team building a loan approval model finds that applicants from one demographic group are approved at a much lower rate, even when controlling for similar financial features. What is the most appropriate next step?

A company wants to build a text summarization feature. Early prototypes sometimes produce fluent but incorrect statements that are not supported by the source text. What is the BEST mitigation strategy to reduce this risk in production?

An organization is adopting AI across multiple departments. They need a governance approach that defines who can approve model deployment, how risk is assessed, and how compliance evidence is maintained. Which artifact best addresses these needs?

Which statement best distinguishes Artificial Intelligence (AI) from traditional deterministic software?

A team wants to measure how well a classification model performs across different probability thresholds. Which metric is most appropriate?

A developer is building an AI feature and wants the application to return a clear message when confidence is low rather than forcing a potentially wrong answer. What is the best practice?

A customer support classifier shows 95% accuracy, but most tickets belong to one category and minority categories are misclassified. Which evaluation approach is most appropriate to understand performance?

A team trains a model and achieves excellent performance on the training set but significantly worse performance on new data. What is the most likely issue and a common mitigation?

A sentiment model was trained on social media text. After deployment to analyze formal customer emails, performance drops. What is the best explanation?

An application uses a generative AI model to answer employee questions based on internal policy documents. Users report that responses sometimes include plausible but incorrect details not found in the documents. Which design pattern best addresses this?

A data scientist is asked to prepare training data for a supervised learning model. Which workflow best aligns with good ML practice?

A hiring-screening model shows lower selection rates for a protected group. The business wants to reduce unfair outcomes while maintaining predictive usefulness. Which action best aligns with responsible AI practices?

A regulated organization must explain individual credit decisions made by an ML model and enable auditability. Which combination is most appropriate?

A retail analyst wants to automatically group customers based on similar purchasing behavior, but there are no existing labels such as "high value" or "low value." Which machine learning approach best fits this goal?

A team is evaluating an AI system for loan approvals and wants to reduce the risk of harm to applicants who are incorrectly denied. Which metric most directly captures this concern?

In a conversational AI project, which artifact is most important to collect early to ensure the assistant understands what users want to accomplish?

A model performs extremely well on training data but poorly on new, unseen data. Which situation is the most likely explanation?

A team building an AI feature wants to ensure they can reproduce predictions and explain changes when a model is updated. Which practice best supports this requirement?

A product team wants to add document question answering over thousands of internal PDFs. Users must receive answers grounded in the documents, and the system should cite sources. Which architecture is most appropriate?

A model is deployed for fraud detection. Over time, legitimate customer behavior changes and the model’s performance degrades. What is the most likely cause?

A healthcare organization wants to use patient records to train a model. They need to minimize privacy risk while still supporting useful analytics. Which approach is most appropriate?

A classifier is trained to detect rare manufacturing defects (about 1% of items). The model achieves 99% accuracy but misses most defects. Which evaluation approach is most appropriate for this scenario?

A team fine-tunes a text model on internal support tickets. After deployment, it sometimes repeats sensitive customer information from training examples. Which mitigation is most appropriate to reduce this risk while keeping the solution practical?

A retail team is evaluating whether a task requires AI. They currently use a fixed set of business rules to apply discounts, and the rules rarely change. Which statement best describes whether AI is needed?

You are building a binary classifier to detect fraudulent transactions. Only 1% of transactions are fraud. Which evaluation metric is generally most appropriate to prioritize during model selection?

A support chatbot must answer questions using only an internal policy document set and must not invent answers. Which approach best fits this requirement?

An organization wants to ensure that only authorized employees can access a dataset containing personal information used for model training. Which control most directly supports this goal?

A data scientist notices that a model performs extremely well on training data but significantly worse on new, unseen data. What is the most likely issue?

A team wants to build an image classifier for identifying damaged products. They only have 400 labeled images. Which strategy is most appropriate to improve results without collecting a massive new dataset?

An AI team needs to operationalize a model so other applications can use it reliably. Which deployment pattern best supports broad integration across systems?

A hiring-screening model shows lower selection rates for a protected group. Before changing the model, what is the best first step in a responsible AI workflow?

A bank must provide clear reasons when a credit decision is denied. Which model choice most directly supports explainability while remaining practical for many tabular credit datasets?

After deployment, a demand-forecasting model’s error increases steadily over several weeks. Training code has not changed. Which action best addresses the most likely root cause?

A retailer wants to explain AI to non-technical stakeholders. Which statement best describes the difference between Artificial Intelligence (AI) and Machine Learning (ML)?

A team built a binary classifier to detect fraudulent transactions. Fraud cases are rare, and the model achieves 99% accuracy by predicting "not fraud" almost always. Which metric is the most appropriate to evaluate performance on the minority (fraud) class?

A product manager wants a conversational assistant to answer customer questions using internal policy documents. What is the most appropriate AI approach?

A data scientist observes that a model performs very well on the training set but significantly worse on unseen validation data. Which issue is most likely occurring?

A bank must provide reasons for loan denials produced by an ML model. Which approach best supports explainability for individual predictions?

An ML pipeline shows excellent validation results, but performance drops sharply after deployment. The input data distribution has shifted due to a change in customer behavior. What is the best next step?

A team is preparing training data for an image classification model. Which practice best helps prevent label noise from harming model quality?

A healthcare organization wants to use patient notes to train a model, but must reduce privacy risk. Which technique is most appropriate as a privacy-preserving measure before model training?

A model is trained to screen job applicants. After deployment, an audit finds that the model systematically ranks one demographic group lower, even when qualifications are similar. Which action best aligns with responsible AI governance?

A team wants to deploy a generative AI assistant for internal use. They are concerned about the model producing plausible but incorrect statements. Which design pattern best reduces this risk in production?