AIF-C01 Study Guide · AI Fundamentals

If you’ve started studying for the AWS Certified AI Practitioner (AIF-C01) exam, you’ve probably noticed that people use “AI,” “ML,” “deep learning,” and “generative AI” as if they were the same thing. They aren’t. Each term describes a different layer of the same field, and mixing them up is one of the fastest ways to misread a scenario-based exam question.

The good news is that the relationship is simple once you see the structure. In this guide, we’ll build that mental model, attach a realistic example to every layer, and map each layer to the AWS services you’re expected to recognize.

The Big Picture: Nested Circles

Picture four nested layers, with each inner layer belonging to the broader one outside it:

┌──────────────────────────────────────────────┐
│ Artificial Intelligence (AI)                 │
│  ┌────────────────────────────────────────┐  │
│  │ Machine Learning (ML)                  │  │
│  │  ┌──────────────────────────────────┐  │  │
│  │  │ Deep Learning (DL)               │  │  │
│  │  │  ┌────────────────────────────┐  │  │  │
│  │  │  │ Generative AI              │  │  │  │
│  │  │  └────────────────────────────┘  │  │  │
│  │  └──────────────────────────────────┘  │  │
│  └────────────────────────────────────────┘  │
└──────────────────────────────────────────────┘

Most modern generative AI systems are built with deep learning, and deep learning is a subset of machine learning, which is itself a subset of AI. The reverse is not true: plenty of AI isn’t machine learning, and plenty of machine learning isn’t deep learning. Generative AI is commonly implemented using deep neural networks, but avoid treating every possible generative system as necessarily deep learning.

Artificial Intelligence: The Broadest Term

Artificial intelligence is any technique that lets a computer perform tasks we associate with human intelligence, such as reasoning, perceiving, deciding, understanding language, or solving problems.

The definition says nothing about how the system does it. That’s why AI includes systems with no learning at all.

Consider a regional pharmacy chain that routes customer emails using a fixed list of rules written by engineers: if the message contains “refill,” send it to the prescriptions queue, and if it contains “insurance,” send it to billing. That is a rules-based system. It automates a human-like task, so it’s AI in the broad sense. But it never improves from data, so it isn’t machine learning.

Remember This: AI is the umbrella. If a question describes automation of a human-like task and says nothing about learning from data, don’t assume ML is involved.

Machine Learning: Learning Patterns From Data

Machine learning is the subset of AI where algorithms learn patterns from data instead of following explicitly programmed rules. You supply examples, the algorithm finds the statistical relationships, and the result is a model you can apply to new data.

Two terms you’ll see constantly:

  • Training: The process of fitting a model to data.
  • Inference: Using the trained model to make predictions on new data.

Return to the pharmacy chain. Instead of keyword rules, the team trains a model on 100,000 past emails that staff already labeled by department. The model learns that the phrase “my card was declined” usually belongs with billing, even though the word “billing” never appears. That is machine learning.

The Three Main Learning Styles

StyleTraining dataWhat the model learnsExample
Supervised learningLabeled (inputs with known answers)Map inputs to known outputsPredicting a delivery’s arrival time from past deliveries
Unsupervised learningUnlabeledHidden structure or groupingsSegmenting customers by purchasing behavior without predefined segments
Reinforcement learningAn agent interacts with an environment and receives feedback or rewardsActions that maximize cumulative rewardA warehouse robot learning efficient routes through trial and error

Within supervised learning, two problem types matter for the exam:

  • Classification predicts a category (spam or not spam, fraudulent or legitimate).
  • Regression predicts a number (a home’s sale price, next month’s energy demand).

You may also see semi-supervised and self-supervised learning. Self-supervised learning matters here because it’s how large language models are commonly pretrained: the model learns to predict missing or next pieces of text from the text itself, so no human labeling is needed for each training example.

Here’s where candidates often get confused: Unsupervised learning isn’t “less accurate” learning. It’s a different problem. If there are no labels and the goal is to find groups, that’s unsupervised, regardless of how sophisticated the algorithm is.

Deep Learning: Neural Networks With Many Layers

Deep learning is a subset of machine learning that uses artificial neural networks with many layers. Each layer transforms the data a little, and stacking layers lets the network learn increasingly abstract features.

Take image recognition. Early layers might detect edges, middle layers combine edges into shapes, and later layers recognize whole objects like “forklift” or “pallet.”

Classical ML often depends on feature engineering, where a person decides which input characteristics matter (for instance, “number of exclamation marks” in an email). Deep learning largely learns its own features from raw data. That’s powerful, but it comes with trade-offs:

FactorClassical ML (e.g., linear regression, decision trees)Deep learning
Data neededCan work with smaller datasetsTypically needs large datasets
ComputeModestOften requires GPUs or specialized accelerators
InterpretabilityOften easier to explainOften harder to explain
Best atStructured/tabular dataImages, audio, text, complex patterns
Common Exam Trap: Don’t assume deep learning is always the best choice. If a scenario describes a small, structured spreadsheet of sales data and a requirement to explain each prediction to auditors, a simpler ML model is often the more sensible answer.

Generative AI: Creating New Content

Generative AI refers to models that create new content, such as text, images, code, audio, or video, rather than only classifying or predicting from existing data. It is built on deep learning, most commonly on the transformer architecture for language.

The models behind it are called foundation models (FMs). They’re pretrained on enormous, broad datasets and can be adapted to many tasks. When the foundation model works primarily with text, it’s a large language model (LLM).

Compare two models at the same company:

  • A predictive (discriminative) model reads a customer review and outputs “negative.”
  • A generative model reads the same review and drafts a polite, personalized response to the customer.

The first picks from a fixed set of answers. The second produces open-ended content.

Why Generative AI Feels Different

  • One model, many tasks. The same foundation model can summarize, translate, classify, and answer questions, guided by prompts.
  • Natural-language interface. You instruct it in plain English rather than writing training code.
  • New risks. Models can produce hallucinations (confident but incorrect output), and you need to think about data privacy, bias, and misuse.

Side-by-Side Comparison

AIMachine learningDeep learningGenerative AI
Core ideaMachines performing human-like tasksLearning patterns from dataMulti-layer neural networksCreating new content
Needs data to learn?Not necessarilyYesYes, usually lotsYes, very large datasets for pretraining
Typical outputDecision or actionPrediction or groupingPrediction from complex dataText, images, code, audio
ExampleRule-based email routerModel predicting which customers will cancelModel detecting defects in factory photosAssistant that drafts maintenance reports

Mapping the Layers to AWS

You don’t need to memorize every service for this topic, but you should know where each layer lives on AWS.

LayerRepresentative AWS offeringsWhat to remember
Machine learning (build your own)Amazon SageMaker AIBuild, train, and deploy custom ML models. This is for teams that want control over the model.
Pretrained AI services (no ML expertise needed)Amazon Rekognition, Amazon Textract, Amazon Comprehend, Amazon TranscribeAPI-based services that use trained models behind the scenes.
Generative AIAmazon BedrockManaged access to foundation models from multiple providers through a single API.

Notice the pattern. AWS offers a spectrum from “call a ready-made API” to “build and train your own model.” Exam scenarios often hinge on how much control, customization, and ML expertise the organization has. For a closer service-by-service comparison, see our guide to our comparison of Amazon Rekognition, Textract, Comprehend, and Transcribe: https://cloudexampro.com/aws-ai-services-compared-rekognition-textract-comprehend-transcribe/.

A Practical Scenario: Same Company, Four Layers

A national home-improvement retailer wants to modernize customer service.

  1. AI (rules): A virtual assistant follows a decision tree: “Press 1 for store hours.”
  2. Machine learning: A model trained on historical orders predicts which customers are likely to return a product.
  3. Deep learning: A model analyzes photos customers upload of damaged products and classifies the type of damage.
  4. Generative AI: An assistant drafts a personalized response to a customer complaint, summarizing the order history and proposing a resolution for an agent to review.

All four are AI. Only the last is generative. Seeing them in one business helps you avoid treating them as competing technologies. In practice, they’re layered.

Common Mistakes and Misconceptions

  • “AI and ML are synonyms.” ML is one way of achieving AI, not the only one.
  • “Generative AI replaced traditional ML.” Predictive ML remains the better and cheaper choice for many tasks, such as forecasting demand or scoring risk.
  • “Deep learning means any complex model.” Deep learning specifically means neural networks with multiple layers.
  • “Bigger models are always better.” Larger models usually cost more and respond more slowly. Fit matters more than size.
  • “Foundation models are trained for one purpose.” Their value is broad pretraining that can be adapted to many tasks.

Exam-Focused Insights for AIF-C01

The AIF-C01 exam guide includes domains on the fundamentals of AI and ML and on the fundamentals of generative AI, so these distinctions are foundational. We can’t predict which topics appear on any given exam, but these skills are well worth building:

  • Spot the clue words. “Labeled historical data” points to supervised learning. “Group similar customers” points to unsupervised. “Reward” and “environment” point to reinforcement learning.
  • Match the output type. A category is classification, a number is regression, and new text or images means generative AI.
  • Separate training from inference. Questions about cost and latency often hinge on which one is being described.
  • Choose the right level of AWS service. Prebuilt API, managed foundation model, or custom model in SageMaker AI.

The easiest way to remember this is to ask two questions of any scenario: “Does it learn from data?” and “Does it create new content?” No to both means rules-based AI. Yes to the first only means ML or deep learning. Yes to both means generative AI.

Key Takeaways

  • AI is the umbrella term. ML, deep learning, and generative AI are progressively narrower subsets.
  • Machine learning learns patterns from data. Supervised, unsupervised, and reinforcement learning differ by how the data and feedback are structured.
  • Deep learning uses multi-layer neural networks and shines on unstructured data such as images, audio, and text.
  • Generative AI creates new content and is powered by foundation models, typically built on deep learning.
  • On AWS, SageMaker AI supports custom ML, the pretrained AI services handle common tasks through APIs, and Amazon Bedrock provides foundation models.

Original Practice Questions and Detailed Answers

These are original, unofficial educational practice questions written by CloudExamPro. They are not actual AWS exam questions.

Question 1 — Rules-based AI vs. machine learning

A hospital’s scheduling team builds a system that routes appointment requests using a fixed list of if-then rules written by developers. The system never changes unless a developer edits the rules. Which statement best describes this system?

A. It is machine learning because it automates a human task.
B. It is artificial intelligence but not machine learning.
C. It is deep learning because it processes many requests.
D. It is generative AI because it produces routing decisions.

Correct answer: B.

  • B is correct. The system performs a human-like task, so it’s AI in the broad sense. It doesn’t learn from data, so it isn’t ML.
  • A is incorrect. Automation alone doesn’t make something ML. ML requires learning patterns from data.
  • C is incorrect. Deep learning requires multi-layer neural networks. Request volume is irrelevant.
  • D is incorrect. Generative AI creates new content. Selecting a route from fixed options isn’t generation.
Question 2 — Supervised regression

A logistics company has five years of delivery records, each labeled with the actual delivery time in minutes. It wants a model that predicts delivery time for new orders. What type of problem is this?

A. Unsupervised clustering
B. Reinforcement learning
C. Supervised regression
D. Supervised classification

Correct answer: C.

  • C is correct. The data is labeled and the target is a continuous number, which is regression.
  • A is incorrect. Clustering applies when there are no labels and the goal is to find groups.
  • B is incorrect. Reinforcement learning uses an agent and rewards, not a historical labeled dataset.
  • D is incorrect. Classification predicts categories, not a numeric value such as minutes.
Question 3 — Unsupervised learning

A streaming service has viewing data for millions of users but no predefined categories. It wants to discover natural groups of users with similar habits. Which approach fits best?

A. Supervised classification
B. Unsupervised learning
C. Supervised regression
D. A rules-based decision tree

Correct answer: B.

  • B is correct. There are no labels, and the goal is to discover structure, which is unsupervised learning (for example, clustering).
  • A is incorrect. Classification needs labeled categories to learn from.
  • C is incorrect. Regression predicts a numeric target, and there’s no target here.
  • D is incorrect. A rules-based tree requires someone to define the groups in advance, which contradicts the discovery goal.
Question 4 — Relationship between the layers

Which statement correctly describes the relationship among AI, ML, deep learning, and generative AI?

A. Machine learning is a subset of deep learning.
B. Generative AI is a separate field with no connection to deep learning.
C. Deep learning is a subset of machine learning, and generative AI is commonly built on deep learning.
D. All four terms mean the same thing.

Correct answer: C.

  • C is correct. Deep learning is a subset of ML, which is a subset of AI. Generative AI models are typically built on deep neural networks such as transformers.
  • A is incorrect. The relationship is reversed. Deep learning sits inside ML.
  • B is incorrect. Generative AI relies heavily on deep learning.
  • D is incorrect. The terms differ in scope and meaning.
Question 5 — Selecting an AWS generative AI service

A furniture retailer wants to automatically draft first-pass product descriptions from a short list of attributes, with editors reviewing each draft. The team has no ML engineers and wants to avoid training a model. Which approach is most appropriate?

A. Train a custom regression model in Amazon SageMaker AI.
B. Use a foundation model through Amazon Bedrock.
C. Write a keyword-matching rules engine.
D. Use Amazon Transcribe to convert the attributes into text.

Correct answer: B.

  • B is correct. Drafting new text is a generative task, and Amazon Bedrock offers managed foundation models with no training required.
  • A is incorrect. Regression predicts numbers, and building a custom model requires the ML expertise the team lacks.
  • C is incorrect. A rules engine can’t produce fluent, varied descriptions.
  • D is incorrect. Transcribe converts speech to text. The inputs here aren’t audio, and it doesn’t generate descriptions.

Frequently Asked Questions

1. Is ChatGPT-style technology the same as AI?

It’s a type of AI, specifically generative AI built on large language models. AI is a much broader category that also includes rules-based systems, predictive ML, and computer vision.

2. Do I need to understand the math behind neural networks for AIF-C01?

No. The exam is aimed at conceptual understanding. You should know what neural networks do and when deep learning makes sense, not derive the algorithms.

3. Is all machine learning “deep learning”?

No. Decision trees, linear regression, and k-means clustering are machine learning but not deep learning, because they don’t use multi-layer neural networks.

4. What’s the difference between a foundation model and an LLM?

A foundation model is a broad, pretrained model adaptable to many tasks and may handle text, images, or other data. An LLM is a foundation model that primarily works with language.

5. When should I choose predictive ML over generative AI?

Choose predictive ML when you need a specific, measurable prediction, such as a churn score or a demand forecast. Choose generative AI when you need new content or flexible language understanding.

Official AWS References

Put Your AIF-C01 Knowledge to the Test

Once these four layers feel clear, test your understanding with our free AIF-C01 diagnostic exam. It’s a low-pressure way to see which fundamentals need more review. When you’re ready for deeper practice, the AIF-C01 Premium practice exams offer original scenario-based questions with detailed explanations. No dumps. No shortcuts.

Take the Free AIF-C01 DiagnosticExplore AIF-C01 Premium

Keep Reading

CloudExamPro creates original educational content to help candidates understand AWS concepts. AWS and its certification programs are trademarks of Amazon.com, Inc. or its affiliates. This article is independent and is not endorsed by AWS.

Leave a Reply

Your email address will not be published. Required fields are marked *

COMING SOON • BLACK FRIDAY MEETS CYBER MONDAY
Something BIG Is Coming! Get ready for an exclusive offer on AWS certification exam preparation. November 27–30, 2026
Starts in--d --h --m --sView Pricing Plans