50 IBM A1000-075: Foundations of AI Practice Questions: Question Bank 2025

A project team is discussing whether their system is using AI, machine learning (ML), or deep learning. Which statement is most accurate?

A customer support manager wants an AI system that can interpret customer questions written in natural language and route them to the right department. Which capability is primarily required?

A team is building an AI assistant and wants a managed service to create a conversational interface that can integrate with business workflows. Which IBM capability best fits this need?

A data scientist notices their classification model performs extremely well on the training set but much worse on new data. What is the most likely issue?

A retail company wants to build a model to predict whether a customer will churn. They have a labeled dataset indicating past churn outcomes. Which learning approach is most appropriate?

A healthcare organization wants an AI solution for analyzing radiology images to highlight suspicious regions. Which type of AI capability is most relevant?

A team is deploying a generative AI solution and wants to reduce the risk of the model exposing sensitive personal information in its responses. Which approach is a best practice?

A team is unsure whether to use accuracy as the primary metric for a fraud-detection model. Only 1% of transactions are fraudulent. Which metric is typically more informative for this scenario?

An enterprise wants to implement a Retrieval-Augmented Generation (RAG) pattern to answer questions using internal policy documents while reducing hallucinations. Which architecture best matches RAG?

A lending model is suspected of disadvantaging a protected group. The organization wants a governance approach that both detects the issue and supports corrective action over time. Which combination is most appropriate?

A product team is deciding whether their new feature uses AI. They want a quick way to describe an AI system that learns from data and improves performance on a task without being explicitly programmed for every rule. Which term best fits this description?

A call center wants a virtual assistant to answer common customer questions using a curated knowledge base (FAQs, policy documents) and provide citations in responses. Which IBM capability best aligns with this need?

A data scientist builds a model that performs extremely well on training data but poorly on new, unseen data. Which issue is MOST likely occurring?

A bank is deploying an AI system to assist with loan decisions. Regulators require clear reasoning for adverse decisions. Which approach BEST supports this requirement?

A team is building a chatbot to handle customer requests that sometimes require account-specific actions (e.g., changing an address). What is the BEST practice to reduce security risk when integrating the assistant with backend systems?

A retailer wants to measure how well a binary classification model identifies fraudulent orders. Fraud is rare, and the team wants a metric that captures how many flagged orders are actually fraud. Which metric is MOST appropriate?

A team fine-tunes a language model on internal documents. After deployment, users report the model sometimes generates plausible but incorrect policy statements. Which architectural change MOST directly reduces this risk while keeping responses tied to approved content?

An AI project team needs to ensure that personal data is used only for the originally stated purpose and not repurposed later without approval. Which governance principle does this MOST closely align with?

A team trains a model and reports 98% accuracy. Later they discover the test set accidentally included many duplicate records from the training set due to a join error. What is the MOST likely consequence of this issue?

A team is asked to assess whether an AI system treats different demographic groups fairly. Which action is the BEST starting point before selecting a fairness metric?

A product team is reviewing AI proposals. One proposal describes a system that “learns by trial and error using rewards and penalties” to optimize actions over time. Which AI learning paradigm is being described?

A support chatbot is giving plausible but incorrect answers about a company’s return policy. The policy is present in an internal document repository. What is the most effective architectural approach to reduce these incorrect answers while keeping responses grounded in the company’s documents?

A data scientist evaluates a binary classifier on a highly imbalanced dataset (1% positive class). Accuracy is 99% but the model rarely identifies positives. Which metric is most appropriate to assess performance on the minority class?

A team builds an AI model to help screen job applicants. During validation, they find the model rejects qualified candidates from a protected group more often than others. What is the BEST next step from an AI ethics and governance perspective?

A developer is comparing AI tasks. Which task is MOST aligned with natural language understanding (NLU) rather than natural language generation (NLG)?

A machine learning pipeline uses one-hot encoding on a categorical feature. In production, a new category value appears that was never seen during training, causing errors and degraded predictions. What is the BEST preventive approach?

An enterprise wants a governed process to track and demonstrate that deployed models meet internal policies (e.g., documentation, approval workflow, ongoing monitoring). Which governance artifact BEST supports auditability across the model lifecycle?

A team fine-tunes a text classification model and observes excellent training performance but significantly worse validation performance. Which issue is MOST likely occurring, and what is an appropriate mitigation?

A healthcare organization plans to use a foundation model to draft patient communications. They must minimize the risk of exposing sensitive information and ensure outputs can be traced to approved sources. Which design is BEST aligned with strong governance and privacy practices?

A company wants to integrate multiple AI services (prompt-based generation, document retrieval, and conversation) into a single application. They also need consistent access control and the ability to swap underlying models without refactoring the entire app. Which architectural principle BEST addresses this requirement?

A team is building an AI-enabled help desk assistant. They want to reduce user confusion by making sure the assistant does not invent answers when it is unsure, and instead either asks clarifying questions or cites verified sources. Which approach best addresses this requirement?

A retail company wants to classify customer support tickets into categories (e.g., billing, returns, technical). They have thousands of historical tickets already labeled by agents. Which type of machine learning problem is this?

A project team wants to identify whether their AI model performs differently across demographic groups (e.g., age ranges) before deploying it. Which evaluation activity most directly supports this goal?

A developer is selecting an IBM Watson service to extract key entities (such as person names, organizations, and locations) from a large set of text documents. Which capability are they primarily looking for?

A data scientist notices that a model has very high accuracy on the training set but significantly lower accuracy on a held-out test set. What is the most likely issue?

A bank is deploying an AI model for credit decisions. Regulators require the bank to explain why a customer was denied credit. Which model choice best supports interpretability while still solving a tabular classification problem?

A team is building a customer-facing chatbot and wants to ensure sensitive data (like account numbers) is not stored unnecessarily in logs and is protected in transit. Which combination best aligns with security and privacy best practices?

An organization plans to use a generative AI model to draft internal policy summaries. They want to reduce the risk of outdated or incorrect information by ensuring responses are based only on the latest approved documents. Which architecture pattern best fits this need?

A model is trained to predict employee attrition. During evaluation, the team discovers that the training dataset includes a feature created using information recorded after the employee left (e.g., exit interview score). Test performance is extremely high but drops sharply in production. What is the root cause?

A healthcare provider wants to deploy a conversational AI assistant. They must demonstrate governance controls including risk assessment, human oversight for critical decisions, and auditability of model behavior over time. Which set of actions best meets these requirements?

A product team wants to explain to non-technical stakeholders why their model sometimes makes incorrect predictions, without diving into math. Which term best describes a technique that provides human-understandable reasons for model outputs?

A team is building a chatbot that must route customer messages to the right department (billing, technical support, or sales). Which Watson capability is the most appropriate to identify the user's intent from text?

You are preparing data for a supervised learning model. Which statement best describes what a "label" is?

A retail company deploys a customer-churn model. Six months later, the model accuracy drops because customer behavior has changed due to a new competitor. What is the most likely cause of this decline?

A bank uses an AI model to recommend credit limits. An auditor asks how the bank can demonstrate accountability and trace decisions back to specific model versions and training data. Which practice best addresses this requirement?

A team builds a multilingual support assistant. They notice the assistant performs well in English but poorly in one regional language. Which is the best next step to improve performance while following good ML practice?

A developer uses an LLM to draft customer emails. Sometimes the model invents order numbers that do not exist. What is the most effective architectural mitigation for this issue?

You are evaluating a binary classifier for fraud detection where only 1% of transactions are fraudulent. Accuracy is 99%, but the model misses many fraud cases. Which metric is most appropriate to prioritize if the goal is to catch as many fraudulent transactions as possible?

A healthcare organization wants to use patient notes to predict readmission risk. The notes contain protected health information (PHI). Which approach best aligns with responsible AI and privacy-by-design while enabling model development?

A team trains a model to predict equipment failure. During evaluation, they discover that a feature called "days_since_last_failure" was calculated using logs that include events occurring after the prediction timestamp. What is the primary issue, and what is the best corrective action?