AWS AI Practitioner Fundamentals of AI and ML Practice Questions (Domain 1, 20%)

Domain 1 is 20% of scored content. It covers AI, ML and deep learning terms, supervised, unsupervised and reinforcement learning, batch and real-time inference, practical use cases and when AI isn't appropriate, the ML lifecycle and MLOps, and model and business metrics. Answer these 10 free questions with no signup, written to AWS's AIF-C01 exam guide v1.2.

Last updated November 11, 2026.

10 Free AIF-C01 Fundamentals of AI and ML Questions with Answers

Question 1: Fundamentals of AI and ML

A car insurer holds four data sources: (a) a claims table in a data warehouse with amounts and dates, (b) photos of vehicle damage, (c) adjusters' free-text notes, (d) a CSV file of policy numbers and premiums. Which combination of sources is unstructured data?

  1. The vehicle damage photos, and the adjusters' free-text notes
  2. The claims table in the warehouse, and the CSV file of policies
  3. The vehicle damage photos, and the claims table in the warehouse
  4. The adjusters' free-text notes, and the CSV file of policies

Correct answer: A. A is correct because photos and free-form notes have no predefined rows and columns, which makes them unstructured, while the claims table and CSV file follow a fixed schema. B is incorrect because both have fixed columns such as amounts, dates and premiums, which makes them structured data. C is incorrect because the photos are unstructured, but the claims table has defined columns and is structured. D is incorrect because the notes are unstructured, but a CSV file of policy numbers and premiums has a fixed schema and is structured. Exam guide: Domain 1 (Fundamentals of AI and ML), Task Statement 1.1.

Question 2: Fundamentals of AI and ML

An insurer is cataloguing its data for AI projects. Match each data source to the type of data it best represents. 1. A table of policyholders with age, postal code and annual premium in fixed columns. 2. Recorded audio of customer phone calls. 3. Daily claim counts for each of the last ten years.

  1. 1 unstructured, 2 time-series, 3 structured tabular
  2. 1 time-series, 2 structured tabular, 3 unstructured
  3. 1 structured tabular, 2 time-series, 3 unstructured
  4. 1 structured tabular, 2 unstructured, 3 time-series

Correct answer: D. D is correct because the policyholder table has defined rows and columns, call recordings have no predefined schema, and daily counts over ten years are ordered measurements over time. A is incorrect because a table with fixed columns is structured data, audio recordings are unstructured, and daily counts over ten years are time-series. B is incorrect because the policyholder table has no time ordering, call recordings have no rows or columns, and daily claim counts are a sequence over time. C is incorrect because call recordings are unstructured media, and daily claim counts are ordered numeric values rather than unstructured data. Exam guide: Domain 1 (Fundamentals of AI and ML), Task Statement 1.1.

Question 3: Fundamentals of AI and ML

A hotel group collects guest reviews that contain written comments and some attached photos. It wants to know which comments mention cleanliness, staff or location so it can route them to the right managers. Which field of AI does this need rely on?

  1. Computer vision, which interprets what appears in the attached photos
  2. Natural language processing, which interprets the meaning of written text
  3. Speech recognition, which converts spoken audio into written text
  4. Reinforcement learning, which learns actions from rewards and penalties

Correct answer: B. B is correct because the routing depends on what the written comments say, and NLP is the field that extracts meaning and topics from text. A is incorrect because the photos are incidental; the need is to understand which topics the written comments mention. C is incorrect because the reviews are already written, so there is no spoken audio to convert. D is incorrect because this is a type of learning rather than a field for understanding text, and the task has no reward signal. Exam guide: Domain 1 (Fundamentals of AI and ML), Task Statement 1.1.

Question 4: Fundamentals of AI and ML

A utility wants to predict tomorrow's electricity demand from three years of hourly meter readings. A new analyst plans to shuffle the rows randomly and treat each reading as an independent record. Which property of this data does that plan ignore?

  1. The readings are unlabelled data, so a person must tag each one before training
  2. The readings are unstructured data, so they must be converted into text before use
  3. The readings are time-series data, so their order and seasonal patterns carry information
  4. The readings are image data, so the analyst needs a computer vision model instead

Correct answer: C. C is correct because time-series data is a sequence of measurements over time, and its order, trends and daily or seasonal cycles are exactly what a demand forecast relies on. A is incorrect because past demand values already serve as the targets a forecast learns from, so no manual tagging is needed. B is incorrect because hourly meter readings are numeric values with timestamps, which is structured data, not unstructured text or media. D is incorrect because the readings are numbers over time, not pictures, so computer vision does not apply. Exam guide: Domain 1 (Fundamentals of AI and ML), Task Statement 1.1.

Question 5: Fundamentals of AI and ML

A dental clinic wants customers to book, change or cancel appointments by talking to a bot on the phone or typing to it on the website. The bot must understand what the customer wants, such as "move my cleaning to Friday", and ask for missing details. Which AWS service fits best?

  1. Amazon Comprehend, which detects the sentiment of each customer's message
  2. Amazon Transcribe, which turns each caller's speech into a written transcript
  3. Amazon Polly, which gives the bot a natural-sounding speaking voice
  4. Amazon Lex, which builds conversational interfaces for voice and text

Correct answer: D. D is correct because Lex recognizes the intent behind what a user says or types and prompts for missing details, which is what an appointment bot needs across phone and web. A is incorrect because knowing whether a message is positive or negative does not let the bot carry out a booking conversation. B is incorrect because a transcript alone does not identify the customer's intent or ask for missing details. C is incorrect because Polly only converts text to speech; it cannot understand what the customer wants or manage the conversation. Exam guide: Domain 1 (Fundamentals of AI and ML), Task Statement 1.2.

Question 6: Fundamentals of AI and ML

A streaming service wants to predict, for each subscriber, whether they will cancel within the next 30 days, so it can target a retention offer. It has two years of subscriber history with known outcomes. Which ML technique fits this need?

  1. Clustering, which groups records without labels
  2. Regression, which predicts a continuous number
  3. Binary classification, which predicts one of two outcomes
  4. Dimensionality reduction, which compresses features

Correct answer: C. C is correct because the target is a yes-or-no outcome (cancels or stays), and labelled history makes this a supervised binary classification problem. A is incorrect because the service has known outcomes and wants a prediction for each subscriber, not unlabelled groupings. B is incorrect because the question asked of each subscriber is whether they cancel, which is a category rather than a continuous number. D is incorrect because reducing the number of features may help preparation, but it does not predict whether a subscriber will cancel. Exam guide: Domain 1 (Fundamentals of AI and ML), Task Statement 1.2.

Question 7: Fundamentals of AI and ML

A small bakery takes about 40 online orders a month and wants to predict next month's flour needs. The owner already estimates this well from a simple spreadsheet of past orders. A consultant proposes building a custom ML forecasting model. What is the best response?

  1. Build the model, since ML forecasts are always more accurate than spreadsheets
  2. Keep the spreadsheet, since the cost of ML would outweigh the small gain
  3. Build the model, but train it on data from large industrial bakeries instead
  4. Replace the spreadsheet with a generative AI chatbot that answers order questions

Correct answer: B. B is correct because AI/ML is not appropriate when a simple method already works and the expected benefit is too small to justify the cost of building and maintaining a model. A is incorrect because ML is not always more accurate, especially with very little data, and the cost would far exceed any benefit at 40 orders a month. C is incorrect because data from very different businesses would not reflect this bakery's demand, and the cost-benefit problem remains. D is incorrect because a chatbot does not address forecasting flour needs and adds cost for a problem that is already solved. Exam guide: Domain 1 (Fundamentals of AI and ML), Task Statement 1.2.

Question 8: Fundamentals of AI and ML

A research institute must run a specific open-weight model with its own custom inference code. Security policy requires it to choose the instance types, control the serving container and keep the model on infrastructure it configures in its VPC. Which option fits best?

  1. Self-host the model on a SageMaker AI endpoint that it configures itself
  2. Call a serverless foundation model through the Amazon Bedrock managed API
  3. Use Amazon Comprehend's pre-trained models for every inference request
  4. Ask business users to query the model from dashboards in Amazon Quick

Correct answer: A. A is correct because a self-hosted SageMaker AI endpoint lets the institute pick instance types, bring its own container and inference code and run inside its VPC, giving the control its policy requires. B is incorrect because a managed API hides the serving infrastructure, so the institute could not choose instances or run its own container. C is incorrect because Comprehend offers fixed NLP capabilities, not a way to host a chosen open-weight model with custom code. D is incorrect because Quick is a workspace for business analysis and dashboards, not a way to host and control a model's serving stack. Exam guide: Domain 1 (Fundamentals of AI and ML), Task Statement 1.3.

Question 9: Fundamentals of AI and ML

A team is testing dozens of training runs for a churn model with different data versions and settings. Two weeks later, no one can say which run produced the best result or how. Which MLOps practice addresses this?

  1. Experiment tracking that logs each run's data, settings and metrics
  2. Data labelling that adds known outcomes to more customer records
  3. Production monitoring that checks the deployed model's live predictions
  4. Batch inference that scores the full customer list every night

Correct answer: A. A is correct because experiment tracking, such as managed MLflow in SageMaker AI, records what went into each run and how it performed, so results can be compared and reproduced. B is incorrect because more labels may help training, but they do not record which run produced which result. C is incorrect because monitoring watches a model in production, while the problem is comparing runs during experimentation. D is incorrect because batch inference is a way to run a model in production; it does not record how experiments were run. Exam guide: Domain 1 (Fundamentals of AI and ML), Task Statement 1.3.

Question 10: Fundamentals of AI and ML

A card issuer's fraud model shows 99% accuracy on test data. Only 1% of transactions are fraudulent, and a review shows the model labels almost every transaction as legitimate. What is the best conclusion?

  1. The model is ready for production, since 99% accuracy exceeds most targets
  2. The model is overfitting, since its accuracy on the test data is too high
  3. The test data is too small, since 1% fraud means there were no fraud cases
  4. Accuracy misleads here; recall on fraud would show it misses most fraud

Correct answer: D. D is correct because with 1% fraud, a model that calls everything legitimate still scores about 99% accuracy, so class-specific metrics such as recall and precision are needed to show it catches almost no fraud. A is incorrect because the model labels almost everything legitimate, so it catches very little fraud, which is the whole purpose of the model. B is incorrect because the high score comes from the class imbalance, not from memorizing training data; the model simply predicts the majority class. C is incorrect because 1% of a test set still contains fraud cases, and the problem is that the model fails to flag them, not that they are missing. Exam guide: Domain 1 (Fundamentals of AI and ML), Task Statement 1.3.

Fundamentals of AI and ML Task Statements in the Exam Guide

TaskExam guide task statementDomainQuestions
1.1Explain basic AI concepts and terminologiesFundamentals of AI and ML28
1.2Identify practical use cases for AIFundamentals of AI and ML28
1.3Describe the AI/ML development lifecycleFundamentals of AI and ML22

More AWS AI Practitioner Practice

FlashGenius is not affiliated with or endorsed by Amazon Web Services. AWS and AWS Certified AI Practitioner are trademarks of Amazon.com, Inc. or its affiliates. These are original practice questions, not questions from the real exam. Your score here is practice accuracy, not an AWS score or a pass prediction.