Fundamentals of AI and ML: 5 AIF-C01 Questions with Full Explanations
Fundamentals of AI and ML is a recurring theme in the AWS AI Practitioner (AIF-C01) exam. It carries about 20% of the exam score. The 5 questions below come with a full explanation of why each option is right or wrong.
AWS AI Practitioner (AIF-C01) · Fundamentals of AI and ML · 5 questions with answers and explanations
📊 Set profile
Exam: AIF-C01 · 65 questions · 90 minutes · passing score 700
This set: 4 advanced, 1 applied · 3 scenario, 1 comparison, 1 concept check · 1 with more than one answer
Topics: AI fundamentals, Amazon Q, Kiro, Amazon SageMaker JumpStart, Time-series forecasting
🧠Before you start
Most items describe a plain business need and ask you to label it: which learning approach, which kind of task, whether ML fits at all. Getting the vocabulary right matters more than math. Wrong options often name a real technique that answers a different question, such as grouping similar records when the need is to predict a future value.
Keep in mind
• Look for a known outcome in the historical data. Labeled examples point toward supervised learning, while rewards and trial and error suggest reinforcement learning.
• Compare training and validation results together: weak results everywhere and strong results only on training data are two different problems with different fixes.
• Keep the nesting of terms straight: generative AI sits inside deep learning, which sits inside ML, which sits inside AI. Having the broader capability does not mean having the narrower ones.
Question 1 of 5
A team is screening proposed projects for machine learning. In which situation is machine learning a poor fit for the business problem?
Show answer
✅ Answer: A
Difficulty: advanced
When machine learning is the right tool
ML learns patterns from examples and produces probabilistic predictions. It shines when the rules are too complex to write by hand and some errors are acceptable.
Good fits
• Complex patterns, such as spotting fraud across hundreds of signals.
• Plenty of relevant historical data to learn from.
• Predictions that improve decisions even if they are sometimes wrong, such as recommendations.
Poor fits
• The answer is fully defined by a known formula or rule, such as calculating sales tax or payroll deductions. Ordinary code gives an exact, explainable, cheap and always-correct answer.
• Little or no data, or a need for 100% deterministic results with no tolerance for error.
Example: a tax calculator with an ML model could return $102.37 when the law gives exactly $100.00; a single line of code would never make that mistake.
Why each option works or fails
• ✓ A. Fixed, fully specified formula (correct): when the correct answer is determined by a known rule, ML only adds cost and error; deterministic code is simpler and exact.
• ✗ B. Complex patterns hard to express as rules: this is the classic case where ML adds value, because it learns the patterns that people cannot write down.
• ✗ C. Large amounts of historical data: plentiful relevant data is what ML models need to learn, so this makes ML a better fit, not a worse one.
• ✗ D. Predictions help and some errors are tolerable: ML predictions are probabilistic, so tolerance for occasional errors is a condition for a good fit.
🎯 Key idea: If a fixed formula already gives the right answer, use code, not ML; ML is for complex patterns learned from data.
Question 2 of 5
An executive says: 'We need AI, not machine learning, and we don't need generative AI because we already have AI.' Which statement corrects this?
Show answer
✅ Answer: D
Difficulty: advanced · Topic: AI fundamentals
AI, ML, deep learning and generative AI are nested
Think of the terms as circles inside each other, not as separate products.
• Artificial intelligence (AI) is the widest field: any technique that lets software perform tasks that normally need human judgment. It includes hand-written rule engines as well as learned models.
• Machine learning (ML) is a subset of AI where the system learns patterns from data instead of being given explicit rules. A churn model trained on past customers is ML.
• Deep learning is a subset of ML that uses multi-layer neural networks, which is what made image, speech and language tasks practical.
• Generative AI (GenAI) is built on deep learning, usually very large pre-trained foundation models, and creates new content such as text, images or code.
Why it matters to an executive
"We already have AI" says nothing about which capability you have. A rules-based FAQ bot is AI but cannot learn; a forecasting model is ML but cannot draft a contract. Each inner circle adds capabilities the outer one does not guarantee.
Why each option works or fails
• ✗ A. Three unrelated technologies: they are not independent; each is a subset of the previous one, and GenAI models are trained with ML techniques.
• ✗ B. ML as rules-only: learning from data instead of hand-coded rules is the defining feature of machine learning, so this reverses the definition.
• ✗ C. GenAI broader than AI: the nesting runs the other way. GenAI is a narrow, recent part of AI, and ML is not a type of GenAI.
• ✓ D. ML inside AI, GenAI inside ML (correct): machine learning is one way of building AI, and generative AI is a deep-learning branch of ML, so "having AI" does not mean you already have ML or GenAI capabilities.
🎯 Key idea: AI ⊃ ML ⊃ deep learning ⊃ generative AI; owning an outer circle does not give you the inner capabilities.
Question 3 of 5
A software development team wants an AI assistant embedded in their IDE that can autonomously plan and implement multi-step coding tasks from a written specification, while also considering a separate generative AI assistant that answers questions and automates tasks across various work activities like code review, troubleshooting, and general Q&A. Which two offerings match the IDE-based spec-driven coding assistant and the broader assistant, respectively?
Show answer
✅ Answer: C
Difficulty: applied · Topic: Amazon Q, Kiro, Amazon SageMaker JumpStart
Kiro: spec-driven agentic IDE
Kiro is an AI-powered development environment (IDE) from AWS built around spec-driven development. Instead of writing code prompt by prompt, you start from a specification:
• Kiro turns a feature request into written requirements and a design
• it breaks the design into an ordered list of implementation tasks
• its agent then carries out those tasks across multiple files, which you review
Amazon Q: a broader generative AI assistant
Amazon Q is AWS's generative AI assistant family used across many work activities. Amazon Q Developer answers AWS and coding questions, suggests and reviews code and helps troubleshoot; Amazon Q Business answers questions using company data.
So the IDE assistant that plans and implements work from a written specification is Kiro, and the general-purpose assistant for Q&A, reviews and troubleshooting is Amazon Q.
Note: Amazon Q Business closed to new customers on 2026-07-31; existing customers can keep using it and AWS points new customers to Amazon Quick, but exam questions may still name Q Business.
Why each option works or fails
• ✗ A. Same product under two names: Kiro and Amazon Q are separate offerings with different focus, one a spec-driven agentic IDE, the other a broad assistant used in many places.
• ✗ B. Roles swapped: this reverses them. The spec-driven, plan-then-implement IDE workflow is Kiro's defining feature, while the broad cross-activity assistant is Amazon Q.
• ✓ C. Kiro for spec-driven coding, Amazon Q for broader tasks (correct): Kiro turns specifications into planned, multi-step implementation inside its IDE, while Amazon Q covers general Q&A, code review, troubleshooting and task automation.
• ✗ D. SageMaker JumpStart for both: JumpStart is a hub of pretrained models and solution templates for building ML applications. It is not a coding assistant or a general work assistant.
🎯 Key idea: Spec-to-implementation agentic coding in an IDE points to Kiro; a general assistant for questions, reviews and troubleshooting points to Amazon Q.
Question 4 of 5
A retailer must predict next quarter's weekly sales per store, including seasonality and holidays, with a range of likely values. Which technique fits?
Show answer
✅ Answer: C
Difficulty: advanced · Topic: Time-series forecasting
What time-series forecasting is
A time series is a sequence of values measured over time, such as weekly sales for each store. Forecasting uses the history of that series to predict future values. Good forecasting models capture:
• trend: sales slowly rising or falling
• seasonality: repeating patterns, such as higher sales every December
• related events: holidays and promotions supplied as extra inputs
Prediction intervals
A single number (for example 1,200 units) hides uncertainty. A prediction interval gives a range, such as "between 1,050 and 1,380 units with 80% probability", often produced as quantiles like p10, p50, p90. Planners use the low end to avoid overstocking and the high end to avoid stock-outs. On AWS, SageMaker Canvas includes time-series forecasting that produces such quantile forecasts without code.
Why each option works or fails
• ✗ A. Sentiment analysis of reviews: sentiment analysis classifies the opinion in text as positive or negative. It does not produce future sales numbers.
• ✗ B. Clustering stores by past volume: clustering groups similar stores together, which can help segmentation, but it does not predict next quarter's weekly sales or give a range for each week.
• ✓ C. Time-series forecasting with prediction intervals (correct): it predicts future weekly sales per store from historical data, accounts for seasonality and holidays, and returns a range of likely values.
• ✗ D. Binary classification: binary classification predicts one of two classes, such as yes or no. The retailer needs numeric values for future weeks, not a two-way label.
🎯 Key idea: Predicting future numeric values over time, with uncertainty ranges, is time-series forecasting; classification and clustering answer different questions.
Question 5 of 5
Which TWO signs indicate a model is underfitting? (Select TWO)
Show answer
✅ Answer: A, E
Difficulty: advanced
Underfitting versus overfitting
• Underfitting: the model is too simple or undertrained to capture the pattern. It performs poorly on training data and poorly on validation/test data. Training and validation scores are both low and close together.
• Overfitting: the model memorizes the training data. It scores very high on training but much lower on validation/test, a large gap.
Example: a straight line fitted to clearly curved data gets 60% on training and 58% on validation, which is underfitting. A very deep tree that gets 99% on training but 70% on validation is overfitting.
Typical fixes for underfitting: a more complex model, better features, longer training, or less regularization.
Why each option works or fails
• ✓ A. Poor validation accuracy as well (correct): an underfit model has not learned the pattern, so it also performs poorly on unseen data.
• ✗ B. Improves only on test data: a model does not learn from the test set, and better test than training results is not a recognized sign of underfitting; it usually points to a small or unrepresentative split.
• ✗ C. Near-perfect training accuracy: shows the model fits the training data very well, the opposite of underfitting and a common warning sign of overfitting.
• ✗ D. Large train-test gap: high training but low test accuracy is the classic sign of overfitting, not underfitting.
• ✓ E. Poor training accuracy (correct): failing even on the data it was trained on is the clearest sign that the model is too simple or undertrained.
🎯 Key idea: Underfitting is poor results everywhere, training included; overfitting is great training results with a big drop on new data.
🔗 Keep practicing on this blog
• Can You Answer These 5 AIF-C01 Questions on Fundamentals of Generative AI?
• All AIF-C01 practice sets
• Every exam on this blog
📚 Keep studying AIF-C01
More AIF-C01 questions: https://hyunhp.github.io/cert-bank/#exam=AIF-C01
Study guide: https://hyunhp.github.io/cert-bank/study.html#exam=AIF-C01
These are original practice questions, not real exam questions, and cert-bank is not affiliated with any certification vendor.
Original practice questions written for this blog, not exam dumps.
Comments
Post a Comment