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

A team wants to explain to business stakeholders what "AI" means in simple terms. Which statement best describes AI at a foundational level?

A healthcare chatbot must understand patient questions and respond with appropriate answers. Which IBM Watson AI service is most directly suited to build the conversational interface?

A retailer wants to predict whether a customer will churn (yes/no) based on historical behavior. What type of machine learning problem is this?

An organization is concerned an AI model might unfairly disadvantage a protected group. Which action is the best example of addressing AI ethics?

A data scientist notices a model performs very well on training data but poorly on new, unseen data. What is the most likely issue?

A company wants to monitor deployed AI models for drift and potential bias over time, and to generate explanations for predictions. Which IBM service is primarily intended for this purpose?

A team is building an NLP solution and has many labeled examples of text with categories (e.g., "billing", "technical support", "account"). Which approach best fits this use case?

A team is selecting evaluation metrics for a fraud detection model where fraudulent transactions are rare. Which metric is generally more informative than overall accuracy in this scenario?

A bank builds a loan approval model and discovers that applicants from a particular neighborhood are denied at a higher rate, largely because the neighborhood correlates with a protected attribute. What is the most appropriate mitigation approach?

A company wants to build an end-to-end AI workflow: prepare data, train models, and then deploy them as an API for applications. Which architecture best aligns with IBM’s typical AI toolchain concepts?

A support team wants an AI assistant that can answer questions from internal policy documents (PDFs) and cite the passages used. They want to reduce hallucinations and keep answers grounded in the documents. Which approach is most appropriate?

A team is building an FAQ chatbot using IBM Watson Assistant. They want the assistant to use a generative model for responses while keeping answers constrained to an approved knowledge base. What is the best practice to meet this requirement?

A data scientist evaluates a binary classifier for fraud detection. False negatives are very costly (missing fraud). Which metric should they prioritize when tuning the model?

A bank wants to ensure its loan approval model does not unfairly disadvantage a protected group. Which action best supports responsible AI practices?

A model performs extremely well on the training set but poorly on the validation set. Which issue is most likely occurring?

You are asked to design an AI solution for classifying incoming customer emails into categories (billing, technical support, cancellations). Labeled examples exist for each category. Which machine learning approach fits best?

A team uses IBM Watson Natural Language Understanding (NLU) to extract entities and sentiment from customer feedback. After deployment, they notice many entities are missed due to domain-specific terminology (product names). What is a practical next step?

A company is building a generative AI system that summarizes meeting transcripts. They must prevent leakage of personally identifiable information (PII) in summaries. Which design control is most appropriate?

A model predicts customer churn. The dataset has 95% non-churn and 5% churn. The model achieves 95% accuracy by predicting 'non-churn' for everyone. Which evaluation approach is most appropriate to reveal the model’s weakness?

A company wants to deploy an AI model into production and satisfy governance requirements. They need to document data lineage, monitor model performance over time, and detect drift. Which combination of practices best addresses these needs?

A team is new to AI and wants a concise way to describe the difference between AI, machine learning, and deep learning. Which statement is MOST accurate?

A chatbot project must route user questions to the right internal policy document, then draft a response citing relevant passages. Which approach best fits this requirement?

You are prototyping a Watson Assistant. Users complain that the assistant frequently asks the same clarifying question even when they already provided the detail earlier in the conversation. What is the MOST likely configuration issue?

A company wants to automatically extract key fields (invoice number, total, due date) from scanned PDF invoices that vary in layout. Which Watson capability best addresses this?

A data scientist trains a classification model with 95% accuracy. In production, it performs poorly because positive cases are rare and the model misses most of them. Which metric would have been MOST helpful during evaluation?

A team notices their model’s training performance keeps improving, but validation performance plateaus and then worsens. What is the MOST likely explanation and best immediate mitigation?

A bank uses an AI model to recommend credit limits. Regulators require an explanation for each decision, and customers must be able to contest outcomes. Which approach best supports this requirement?

A healthcare AI proof-of-concept uses patient notes that include names and addresses. The team wants to reduce privacy risk before training. Which action is MOST appropriate?

You are designing an enterprise Q&A assistant that must avoid answering when it cannot find reliable supporting content in company-approved sources. Which design choice best reduces hallucinations?

A model is trained on historical hiring decisions. After deployment, it disproportionately rejects applicants from a protected group. The training data reflects past biased decisions. What is the MOST appropriate next step?

A customer support team wants to automatically route incoming emails into categories like "Billing", "Technical Issue", and "Cancellation". They have several hundred labeled examples per category and the content is plain text. Which approach is most appropriate?

A team is piloting a chatbot and wants to reduce the risk of the model generating harmful or off-topic responses. Which control is the most practical first step for a foundational implementation?

You need an IBM Watson service to extract entities, keywords, sentiment, and categories from customer feedback text. Which Watson capability is best suited?

A project team is debating whether their solution is AI or traditional automation. Which statement best describes a key difference between machine learning and rule-based systems?

A bank trains a loan-approval model and notices 96% accuracy. However, only 4% of applicants are approved in the historical data. Which metric is most appropriate to evaluate whether the model is actually identifying approvals correctly?

A team built a classifier that performs well in testing but fails after deployment because customer behavior changed over time. What is the most likely issue?

A healthcare application uses a model to flag high-risk patients. Clinicians require an explanation of which factors contributed to each prediction. Which design choice best addresses this requirement?

A team wants to build a voice-enabled virtual assistant. Users will speak questions, the system will interpret the audio into text, then generate a reply. Which combination of Watson services best fits the speech portion of this flow?

A retail company wants to personalize product recommendations. They have user purchase history but no explicit "like/dislike" labels. Which approach is most suitable to start with?

A team is training a model to detect defects in manufacturing. They accidentally include a feature indicating which inspector reviewed the item, and that inspector tends to work only on a specific production line with higher defect rates. The model performs extremely well in validation but fails in a different plant. What is the most likely root cause?

A team wants to quickly add a conversational interface to answer common employee HR questions (benefits, holiday policy) without building an NLP model from scratch. Which IBM Watson service is the best fit?

Which statement best describes the difference between AI, machine learning (ML), and deep learning (DL)?

A developer reports that a classification model has 98% accuracy, but users complain it rarely detects the positive class (a rare event). Which metric is most important to review to validate this concern?

A retail company wants to understand customer sentiment about its products from thousands of online reviews and route negative reviews to support. Which IBM Watson capability is most appropriate?

A team trains a model in a notebook and gets strong results, but performance drops significantly after deployment. Data scientists suspect the incoming production data no longer matches the training data distribution. What is the most likely issue?

A bank wants to use AI for loan approvals but must ensure the system can be explained to regulators and applicants. Which approach best supports this requirement?

A project uses human-labeled images for training. During review, the team finds that one labeler consistently marks certain objects differently from everyone else. What is the best next step to improve model reliability?

A team needs to deploy a trained model as a REST endpoint and manage versions so applications can call it consistently. Which IBM component is commonly used for model deployment and serving?

A hiring model shows lower selection rates for a protected group. The team wants to evaluate whether outcomes differ unfairly between groups using a quantitative fairness metric. Which metric directly compares positive outcome rates across groups?

A company wants to operationalize governance by continuously monitoring deployed models for accuracy and bias, capturing model metadata, and supporting explainability dashboards for stakeholders. Which IBM capability best aligns with these goals?