50 IBM A1000-047 - Assessment: Foundations of AI Practice Questions: Question Bank 2025

A customer support team wants an AI system that can answer common questions by retrieving the most relevant passages from product manuals. Which approach best fits this requirement?

A team is deciding between supervised and unsupervised learning for a project. They have thousands of past loan applications with an outcome label (approved/denied). Which method is most appropriate and why?

A product owner asks why an AI chatbot sometimes gives confident but incorrect answers. What is the most accurate explanation of this behavior for large language models?

An organization wants to reduce the risk of exposing sensitive personal data when training an AI model. Which practice is the BEST first step?

A data scientist reports that a classification model has 95% accuracy, but the business says it fails to catch rare fraud cases. The dataset is highly imbalanced (fraud is 1%). Which metric is typically more informative than accuracy in this scenario?

A team uses IBM Watson services to extract key entities and sentiment from customer emails. They want a managed capability that analyzes unstructured text for entities, keywords, categories, and sentiment. Which IBM Watson capability is the best fit?

A machine learning pipeline shows strong training performance but significantly worse test performance. Which issue is MOST likely, and what is the best general remedy?

An enterprise wants to deploy an AI assistant. Security requires that only authorized employees can access it and that prompts/responses be logged for audit. Which design choice best supports these requirements?

A company builds a generative AI feature that summarizes internal HR policies. During testing, the model sometimes cites policies that do not exist. The team already uses retrieval over an approved document repository. What is the MOST effective next step to reduce fabricated citations?

A team is training a model to predict equipment failure. They accidentally include a feature that is computed using information that becomes available only AFTER a failure occurs. The model performs extremely well in validation but fails in production. What is the root cause?

A support team wants to automatically route incoming emails to the correct department (billing, technical, sales). They have historical emails labeled by department. Which approach best fits this problem?

Which statement best describes a key difference between Artificial Intelligence (AI) and Machine Learning (ML)?

A team is building an AI-powered chatbot and wants to add natural language understanding without training models from scratch. Which IBM capability is most appropriate to start with?

A bank trained a loan-approval model and reports 94% accuracy. However, only 3% of applicants are truly high-risk (the positive class). Which additional metric is most important to evaluate whether the model is actually identifying high-risk applicants?

A product team is preparing training data for a customer sentiment model. They want to reduce bias and improve reliability of labels. Which practice is most appropriate?

A team built a proof-of-concept model in a notebook and now needs an API endpoint so applications can send data and get predictions. Which pattern is the best next step?

A model performs well in testing but degrades after deployment because customer behavior changes over time. What is the most likely issue and recommended mitigation?

A retail company wants to use a foundation model to generate product descriptions. They are concerned that prompts could include confidential supplier terms, and generated text might expose sensitive content. Which governance control is most appropriate to reduce risk?

A healthcare organization builds a model to predict readmission risk. During evaluation, they discover the model relies heavily on a feature that directly encodes whether the patient was previously flagged for extra care by clinicians—information that would not be available at decision time for new patients. What is the most accurate diagnosis of the problem?

A company wants to provide explanations for individual credit decisions to meet regulatory requirements. Which approach best supports per-decision interpretability while maintaining a strong governance posture?

A retail team wants to add an AI-powered search bar to their website that answers questions using only the company’s product manuals and policies. They want to reduce the risk of the model inventing details. Which approach is MOST appropriate?

A data scientist evaluates a binary classifier and sees very high accuracy, but the dataset contains 98% negative cases and 2% positive cases. Which metric is MOST useful to understand how well the model identifies the positive class?

A team wants to extract person names, organizations, and locations from customer emails to route issues faster. Which IBM Watson capability best matches this need?

A bank is creating an AI system to help loan officers. The bank must ensure decisions can be explained to customers and regulators. Which model choice is MOST aligned with this requirement?

A team deploys a chatbot and notices users are receiving answers that include outdated policy text. The knowledge base was updated last week. What is the MOST likely cause in a RAG-based system?

A healthcare organization wants to use patient notes to train an NLP model. Regulations require minimizing exposure of personal identifiers and limiting access to sensitive data. What is the BEST first step before model training?

A team must choose between training from scratch and transfer learning for an image classification task with only a small labeled dataset. Which approach is MOST appropriate and why?

An AI assistant is used internally to draft HR policy responses. Leadership wants governance controls to ensure the assistant’s outputs align with company policy and can be audited. Which combination BEST supports this goal?

A company wants to integrate multiple AI capabilities—speech transcription, language understanding, and a conversational interface—into a single customer support workflow. Which architectural approach is MOST appropriate?

A model shows excellent performance in testing but performs poorly after deployment to a new region with different customer behavior. What is the MOST likely explanation and BEST next step?

A help-desk team wants to automatically route incoming emails to the correct department. They have a labeled history of emails with the correct department assigned. Which AI approach best fits this need?

A team is building an AI solution using an IBM-managed service and wants to reduce operational overhead for model hosting and scaling. Which option best matches this goal?

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

An organization wants to ensure that an AI system’s decisions can be explained to non-technical stakeholders (for example, why a loan was denied). Which requirement does this address most directly?

A retail company wants to predict next month’s sales for each store using past sales values and seasonal trends. What type of machine learning problem is this?

A team is using a pre-trained natural language model for sentiment analysis. Their domain includes specialized product slang that the model misinterprets. What is the best next step to improve performance?

A business analyst compares two models: Model A has higher overall accuracy, but Model B has much better recall for the rare “fraud” class. If the priority is to catch as many fraudulent transactions as possible, which model is preferable?

A team is deploying an AI-powered document processing solution. They need an architecture that supports continuous improvement when new document layouts appear. Which approach is most appropriate?

A bank uses an AI model to recommend credit limits. During an internal audit, they discover the model gives systematically lower limits to a protected group even when controlling for income and repayment history. What is the most appropriate first action?

A company wants to provide answers from its internal policy documents using a generative AI assistant. They are concerned about the assistant hallucinating unsupported claims. Which design best reduces hallucinations while keeping answers grounded in company content?

A team is building an AI proof-of-concept and needs a clear way to describe the overall AI workflow from data collection through model use in production. Which concept best represents this end-to-end workflow?

A customer support chatbot must hand off to a human agent when it is unsure. Which chatbot design element most directly enables this behavior?

In a supervised learning project predicting whether an insurance claim is fraudulent, what is the "label"?

A data scientist reports 98% accuracy on a model that predicts a rare disease affecting 1% of patients. What is the most likely issue with relying on accuracy alone?

A company wants to reduce the risk of exposing customer PII while still training an ML model on customer behavior. Which approach is the best practice?

A Watson Assistant bot is not matching user messages to the expected intent in production, even though it worked in testing. Which troubleshooting step is most appropriate first?

A team is building a text classification model and notices it performs extremely well on training data but poorly on new data. Which term best describes this issue?

A business stakeholder asks whether the model treats two demographic groups differently for loan approvals. What is the most appropriate initial analysis to perform?

You must design an architecture where an application sends customer chat transcripts to be analyzed for sentiment and key phrases, and then stores results for dashboards. Which design is most appropriate?

A model used for predicting equipment failures shows a steady drop in performance over several months, even though the code has not changed. Sensors were recalibrated and operating conditions evolved. What is the most likely cause and the best response?