50 IBM A1000-050 - Assessment: Foundations of AI Practice Questions: Question Bank 2025

A team is explaining how an AI system improves over time as it is exposed to more labeled examples. Which concept are they describing?

A retailer wants to predict next month's sales revenue using historical monthly sales values. Which type of machine learning task best fits this requirement?

A bank wants to automatically flag unusual credit card transactions for review, even when there are few confirmed fraud examples. Which approach is most appropriate?

A company wants to reduce the risk of exposing customer data when using AI to generate insights. Which practice best supports responsible AI use?

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

A healthcare team is building a model to classify X-rays as 'normal' or 'needs review.' Only 5% of the images are 'needs review.' Accuracy is 95% even when the model predicts 'normal' for every image. Which metric is most useful to evaluate performance on the minority class?

A customer support center wants a virtual assistant to answer common questions using internal policy documents and provide citations to the specific paragraphs used. Which design best supports this requirement?

An AI team is asked to explain a model's decision to a regulator. Which approach best supports explainability in a governance process?

A team trains a model to predict employee attrition. They unintentionally include a feature that directly encodes the outcome (e.g., 'exit interview scheduled'). The model shows very high performance during evaluation but fails in production. What is the primary problem?

A company deploys an AI model for loan approvals. After deployment, approval rates for a protected group drop significantly compared to the pre-deployment baseline. Which first action best aligns with responsible AI governance?

An insurance company wants to quickly explain to business stakeholders the difference between AI, machine learning (ML), and deep learning. Which statement is MOST accurate?

A retailer is building a model to predict whether a customer will churn (yes/no). Which type of ML problem is this?

A team plans to use customer service chat transcripts to train an AI assistant. Which action BEST supports responsible AI by protecting customer privacy?

A data scientist notices a classification model has 95% accuracy, but it performs poorly in identifying a rare but critical class (e.g., fraudulent transactions). Which evaluation metric is MOST appropriate to focus on improving detection of the rare class?

A bank wants to use AI to help call center agents during live calls by suggesting next-best actions and summarizing key customer issues. Which AI capability is MOST directly aligned with this requirement?

A team trains a model and sees excellent performance on training data but significantly worse performance on new, unseen data. What is the MOST likely issue?

A healthcare organization wants to deploy an AI model that helps prioritize patient outreach. They must demonstrate the model is explainable to clinicians and auditors. Which approach BEST supports this requirement?

A company is designing an AI solution for document processing: ingest PDFs, extract key fields (invoice number, total, vendor), and route exceptions to humans. Which architecture is MOST appropriate?

A model is used to recommend loan approvals. Post-deployment monitoring shows approval rates differ significantly across protected groups, even when applicants have similar financial profiles. What is the BEST next step from an AI governance perspective?

A team built a sentiment analysis model and it performs well in testing. After deployment, performance steadily degrades as new slang and product names appear. Which practice BEST addresses this issue long-term?

An operations team wants an AI system to flag unusual spikes in website traffic without having labeled examples of “bad” behavior. Which machine learning approach best fits this requirement?

A product manager asks why a chatbot sometimes gives confident but incorrect answers that are not present in the company knowledge base. What is the best term for this behavior in generative AI systems?

A retail company wants to quickly demonstrate business value from AI by starting with a small, low-risk use case. Which approach is considered a best practice?

A data scientist reports a model has 98% accuracy predicting a rare fraud event, but fraud occurs in only 1% of transactions. Which evaluation metric is most appropriate to understand performance on the minority class?

A team is building a loan approval model. They discover historical training data reflects past bias against a protected group. Which action best supports ethical AI while maintaining model usefulness?

A team builds a model that performs well in development but degrades after deployment because customer behavior changes over time. What is the most likely issue and recommended response?

A support organization wants to use generative AI to answer employees’ questions using internal policy documents, while reducing the risk of fabricated answers. Which architecture pattern is most appropriate?

During model training, a team accidentally includes a feature that is only known after the prediction time (e.g., “chargeback filed” when predicting fraud at purchase). The model shows unusually high validation results. What is the most likely cause?

A healthcare organization wants to deploy an AI model that impacts patient triage decisions. Which governance practice is MOST important to satisfy accountability and auditability requirements?

A company is building a sentiment classifier and notices the model performs significantly worse on new product categories that were not well represented in training data. Which strategy best improves generalization to these underrepresented categories?

A retail company wants to quickly categorize incoming customer emails into topics (returns, shipping, billing) but has very limited labeled data. Which approach is most appropriate to start with?

In a supervised learning project, the training accuracy is high but the test accuracy is much lower. Which issue is most likely occurring?

A team is selecting an AI technique to transcribe recorded customer calls into text. Which AI capability best matches this requirement?

Which statement best describes the difference between AI, machine learning, and deep learning?

A bank wants an AI model to assist with loan approvals. Regulators require the bank to provide a clear, human-understandable reason for each decision. Which model choice best aligns with this requirement?

A data scientist builds a model to predict churn. They standardize numerical features by calculating mean and standard deviation using the full dataset (train + test) before splitting. Test performance looks unusually strong. What is the most likely problem?

A hospital wants to triage incoming patient messages. Missing an urgent case is far more costly than flagging some non-urgent cases as urgent. Which evaluation focus is most appropriate?

A retailer is deploying an AI system to recommend products. The business wants to detect and reduce systematic under-recommendation of products from small, minority-owned suppliers. Which practice best addresses this goal?

A credit risk model shows strong performance overall, but its error rate is significantly higher for one protected group. The organization wants to improve fairness while keeping performance acceptable. Which approach is most appropriate?

A company wants to use a large language model (LLM) to draft responses to customer inquiries. A security review finds the system sometimes includes sensitive customer data from internal documents in its outputs. Which architecture pattern best reduces this risk while still enabling useful answers?

A product team wants an AI assistant that can answer questions using only the company’s internal policy documents and should cite sources from those documents. Which approach best fits this requirement?

A bank builds a loan approval model. The model performs well overall but is found to reject a higher proportion of applicants from a protected group, even when their financial profiles are similar. What should the team do FIRST?

Which statement best describes the difference between supervised and unsupervised learning?

A team trains a classification model and gets 95% accuracy. In production, only 2% of cases are the positive class and missing positives is costly. Which metric is MOST appropriate to focus on?

A retailer wants to forecast next week’s demand for each store using past sales and promotions. What type of problem is this?

An AI system flags unusual credit card transactions. Fraud is rare, but the business wants to catch new fraud patterns that were not seen during training. Which approach is MOST suitable?

A team notices their deployed model’s performance degrades over time because customer behavior changes and new products are introduced. What phenomenon BEST explains this and what is a common mitigation?

A healthcare organization wants to use a third-party generative AI service to summarize clinician notes. The notes contain highly sensitive personal data. Which governance control is MOST important to implement before use?

A company is building an AI solution and wants stakeholders to understand how the model makes decisions for individual predictions, not just overall feature importance. Which technique best supports this need?

A team deploys an LLM-based chatbot connected to internal systems (e.g., ticketing and HR). During testing, a user prompt causes the chatbot to reveal confidential data by overriding system instructions. What is the MOST appropriate architectural mitigation?