Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) are transforming how businesses solve problems, from recommending what film to watch next to diagnosing diseases from medical scans. For the AIF-C01 exam, you need to understand the differences between these three concepts and how they relate to each other, not just as buzzwords but as a clear hierarchy of capability and complexity.
Jump to a section
A simple way to picture Introduction to AI and ML on AWS
A cookbook is a collection of recipes, each a precise list of ingredients and steps to create a specific dish. A chef follows the recipe without deviation; the result is predictable and repeatable. This is traditional programming: you give the computer exact instructions (the recipe), and it follows them to produce the output.
Artificial Intelligence is like cooking itself — the broad discipline of creating machines that can perform tasks that normally require human intelligence, like tasting and adjusting seasoning. Machine Learning is a specific technique within AI, akin to a chef tasting a dish, learning that it needs more salt, and remembering that adjustment for next time. Instead of following a fixed recipe, the chef (the ML model) learns from experience (data) and improves its performance (the dish) with each attempt.
Deep Learning is a more advanced version of this, like a master chef who can deconstruct a flavour profile into hundreds of subtle components — the acidity of a tomato, the umami of a mushroom — and understand how they interact. Deep learning uses artificial neural networks with many layers (hence 'deep') to process vast amounts of data and find incredibly complex patterns, just as the master chef understands the intricate chemistry of cooking. On AWS, you don't need to build your own kitchen; you use managed services like Amazon SageMaker to orchestrate the whole process, from preparing ingredients (data) to serving the final dish (deploying a model).
Let's start with the broadest term: Artificial Intelligence (AI). AI is the entire field of computer science dedicated to creating systems that can perform tasks that would normally require human intelligence. This includes things like understanding natural language (like Alexa or Siri), recognising objects in images (like tagging photos on social media), playing strategic games (like chess or Go), and making decisions based on data. AI is an umbrella term; it covers everything from a simple rule-based system (like a thermostat that turns on the heat when the temperature drops) to advanced systems that learn and adapt.
Under the AI umbrella sits Machine Learning (ML). ML is a subset of AI. Instead of being explicitly programmed with rules for every possible scenario (e.g., 'if the image has a tail and four legs, it is a dog'), an ML system learns patterns from data. You feed it thousands of labelled images of dogs and cats, and it figures out the distinguishing features on its own. The key is that the model is not explicitly told 'look for floppy ears' — it discovers that pattern from the data. This is incredibly powerful for tasks where writing explicit rules is impossible, like predicting which customers will churn or translating text between languages.
Within ML, there are different types of learning:
Supervised Learning: You show the model labelled data (e.g., 'this email is spam', 'this email is not spam'). The model learns the mapping from input to output and can then predict labels for new, unlabelled data. This is the most common type of ML.
Unsupervised Learning: You give the model data without labels and ask it to find hidden patterns or groupings. For example, a retailer might use it to segment customers into groups based on buying behaviour, without telling the model what the groups are.
Reinforcement Learning: The model learns by interacting with an environment, performing actions, and receiving rewards or penalties. It learns a strategy (policy) to maximise cumulative reward over time. This is how AI masters games like Go or controls a robot arm.
Now, Deep Learning (DL) is a further subset of ML. It uses artificial neural networks with many layers (hence 'deep') to model complex, non-linear relationships in data. A neural network is inspired by the structure of the human brain, with interconnected 'neurons' organised in layers: an input layer, one or more hidden layers, and an output layer. Deep learning excels at tasks like image recognition, speech recognition, and natural language processing because it can automatically learn hierarchical features. For image recognition, the first layer might detect edges, the next layer shapes, the next layer objects, and so on.
The relationship is a series of concentric circles: AI is the broadest, ML is inside AI, and DL is inside ML. Every DL system is ML, and every ML system is AI, but not all AI is ML (a simple rule-based system is AI but not ML), and not all ML is DL (a linear regression model is ML but not DL).
On AWS, this hierarchy maps to services. For traditional AI without ML, you might use Amazon Lex for chatbots (rule-based conversation). For ML, you use Amazon SageMaker to build, train, and deploy models. For DL specifically, SageMaker supports deep learning frameworks like TensorFlow and PyTorch, and services like Amazon Rekognition (image/video analysis) and Amazon Polly (text-to-speech) are pre-built DL applications.
Define the Problem and Identify It as an AI Task
The first step is recognising that a business challenge (e.g., 'predict equipment failures before they happen') can be solved using AI techniques. This sets the stage for deciding whether a simple rule-based AI system or a more advanced ML/DL solution is appropriate.
Gather and Prepare Data (Data Engineering)
Collect historical data (e.g., sensor readings, failure logs) from sources like Amazon S3 or databases. Clean the data to handle missing values, remove duplicates, and format it consistently. This step is essential because the quality of the data determines the quality of the model. AWS Glue and SageMaker Data Wrangler can simplify this.
Choose Between Traditional ML and Deep Learning
Based on the data size, complexity of the problem, and available compute resources, decide whether to use a traditional ML algorithm (like Random Forest for tabular data) or a deep learning neural network (for image or text data). This directly maps to the AI/ML/DL hierarchy: traditional ML is still ML, DL is a specialised subset.
Train and Evaluate the Model
Use Amazon SageMaker to select an algorithm, split data into training and testing sets, train the model, and evaluate its performance using metrics like accuracy or F1-score. This is the core of ML practice: the model learns patterns from the training data and is validated on unseen test data.
Deploy and Monitor the Model
Deploy the trained model to a SageMaker endpoint so it can serve real-time predictions. Set up monitoring with SageMaker Model Monitor to detect data drift or concept drift. Over time, the model may need retraining with new data, completing the ML lifecycle. This step shows that ML is not a one-off task but a continuous process.
Consider a real-world scenario: a large e-commerce company like 'ShopFast' wants to improve its product recommendation system to increase sales. An IT professional (in this case, a data scientist or ML engineer) working with AWS would approach this problem using AI, ML, and potentially DL.
First, the IT professional defines the business problem: 'We want to recommend products to customers that they are likely to buy.' This is an AI problem. They could solve it with a simple rule-based system (e.g., 'recommend the most popular items'), but that wouldn't be personalised. To deliver personalised recommendations, they turn to ML.
The professional uses Amazon SageMaker, AWS's primary ML service. The steps they follow are:
Data Collection: They gather historical data: what customers bought, what they viewed, their ratings, and account details. This data is stored in Amazon S3 (Simple Storage Service).
Data Preparation: They clean and transform the data. They label the data for supervised learning: a 'purchase' is the target variable (what you want to predict), and customer attributes are the features (input). They might use AWS Glue for this ETL (Extract, Transform, Load) process.
Model Training: They choose an ML algorithm. For recommendations, a common choice is the 'Factorization Machines' algorithm or a collaborative filtering approach. They train the model on the historical data. The model learns patterns like 'customers who bought a tent also bought a sleeping bag'.
Model Evaluation: They split the data into training and test sets. They evaluate the model's performance on the test set using metrics like accuracy or RMSE (Root Mean Square Error). They iterate by trying different algorithms or tuning hyperparameters (settings that control the learning process).
Model Deployment: They deploy the trained model as a real-time endpoint using SageMaker hosting. When a customer visits the website, a request is sent to the endpoint with the customer's ID, and the model returns a list of recommended products.
Monitoring and Retraining: They use SageMaker Model Monitor to track data drift (when the incoming data starts to look different from the training data) and retrain the model periodically to keep it accurate.
If the team wanted to improve recommendations further, they might use Deep Learning. For example, they could build a neural network that analyses the actual text of customer reviews (natural language processing) or the images of products the customer viewed (computer vision). Services like Amazon Rekognition (for image analysis) or Amazon Comprehend (for text analysis) are pre-trained DL models they can call via an API, or they can build their own custom DL model using SageMaker with deep learning frameworks.
The IT professional does not need to manage servers for any of this; AWS handles the infrastructure, scaling, and security. They focus on the data, the model, and the business problem.
The AIF-C01 exam tests your understanding of the definitions and relationships between AI, ML, and DL under objective 1.1. Expect 3-5 questions directly on this topic, typically scenario-based or definition-based. Here is exactly what you need to know.
Key concepts to memorise:
AI is the broadest field. Any system that shows intelligent behaviour (rule-based, ML, or DL) is AI.
ML is a subset of AI where systems learn from data without being explicitly programmed for every rule.
DL is a subset of ML that uses multi-layered neural networks to learn complex patterns.
The relationship is: AI > ML > DL. Every DL system is ML and AI, but not vice versa.
Exam question types: - 'Which of the following is an example of machine learning but NOT deep learning?' Correct answer: a linear regression model or a decision tree. Traps: they might list a neural network (which IS deep learning) or a simple if-then rule (which is AI, but not ML). - 'A system that uses a neural network to improve its accuracy over time is an example of which technology?' Correct answer: deep learning. Trap: they might say 'AI' (too broad) or 'rule-based system' (wrong). - 'What is the primary difference between supervised and unsupervised learning?' Correct answer: labelled vs unlabelled data. Traps: they might confuse it with reinforcement learning.
Common trap patterns:
Trap: Equating 'AI' with 'ML'. Remember: AI includes non-ML systems. A calculator is AI, but not ML. The exam loves to test this.
Trap: Thinking all neural networks are deep learning. Shallow neural networks (with only one hidden layer) are considered ML but not DL. DL requires multiple hidden layers (typically more than 2).
Trap: Confusing the level of complexity. Deep learning is not always better; it requires huge amounts of data and compute. The exam may ask when to choose ML over DL.
Trap: Misunderstanding the relationship. They might give a Venn diagram or a list and ask you to identify the correct hierarchical relationship.
Key definitions to memorise:
Artificial Intelligence: Computer systems that perform tasks normally requiring human intelligence.
Machine Learning: A type of AI that gives computers the ability to learn from data without explicit programming.
Deep Learning: A type of ML that uses multi-layered neural networks to learn from vast amounts of data.
Supervised Learning: Learning from labelled data.
Unsupervised Learning: Finding patterns in unlabelled data.
Reinforcement Learning: Learning through trial and error with rewards.
For each question, read the scenario carefully. Identify if the system uses explicit rules (AI, not ML), learns from labelled data (supervised ML), or uses a multi-layer neural network (DL). The exam will also test your ability to match AWS services to these categories: Amazon Rekognition (DL), Amazon Lex (AI, can be ML), Amazon SageMaker (ML/DL platform).
Artificial Intelligence is the broadest discipline; Machine Learning is a subset of AI; Deep Learning is a subset of Machine Learning.
Machine Learning systems learn patterns from data instead of following explicit, human-written rules for every scenario.
Deep Learning uses multi-layered neural networks to automatically discover complex, hierarchical patterns in vast amounts of data.
Supervised learning uses labelled data (input-output pairs), unsupervised learning finds hidden patterns in unlabelled data, and reinforcement learning learns through trial-and-error with rewards.
On AWS, Amazon SageMaker is the primary service for building, training, and deploying both Machine Learning and Deep Learning models.
All Deep Learning is Machine Learning, and all Machine Learning is Artificial Intelligence, but not all AI is ML (rule-based systems are AI but not ML).
These come up on the exam all the time. Here's how to tell them apart.
Artificial Intelligence
Broad field encompassing all smart computer systems
Includes both rule-based systems and learning systems
Example: a chess program with fixed strategies
Machine Learning
Subset of AI that learns patterns from data
Requires data to improve performance over time
Example: a chess program that learns from thousands of past games
Supervised Learning
Uses labelled data (input-output pairs) for training
Goal: predict a label or value for new data
Example: predicting house prices from historical sales data
Unsupervised Learning
Uses unlabelled data to find hidden patterns or groupings
Goal: discover structure in data (clusters, associations)
Example: grouping customers by purchase behaviour without predefined categories
Traditional Machine Learning
Uses algorithms like linear regression, decision trees, SVMs
Works well with small to medium-sized datasets
Typically requires feature engineering by humans
Deep Learning
Uses multi-layered neural networks
Excels with very large datasets (millions of examples)
Automatically learns features from raw data (e.g., pixels)
Mistake
AI and Machine Learning are the same thing; you can use the terms interchangeably.
Correct
AI is the broad field of creating intelligent systems. ML is a specific subset of AI where systems learn from data. All ML is AI, but not all AI is ML (e.g., a rule-based chess program is AI, but not ML).
This mistake is common because marketing and media often use 'AI' loosely to refer to any smart feature, even if it is just a simple algorithm. The distinction is crucial for the exam.
Mistake
Deep learning is always better than traditional machine learning for any task.
Correct
Deep learning requires massive amounts of data and computational power. For small datasets or simpler problems, traditional ML algorithms (like linear regression or decision trees) often perform better and are more interpretable. The choice depends on the problem, not just the trend.
Beginners hear 'Deep Learning = advanced = best' and assume it is the universal solution. The exam tests your ability to choose the right tool for the job, and overusing DL is a common trap.
Mistake
I don't need to understand the data because the ML model will figure everything out automatically.
Correct
Garbage in, garbage out. The quality and relevance of the data directly determine the model's performance. If you feed an ML model biased, missing, or irrelevant data, the model will learn incorrect or harmful patterns. Data preparation and understanding is a critical step.
This misconception arises from the intuitive feel of 'the machine is smart, it will fix it'. Real-world experience shows that data cleaning and feature engineering are the most time-consuming parts of an ML project, and AWS services like SageMaker help, but they do not replace human judgment on data quality.
Mistake
Once a machine learning model is trained and deployed, it stays accurate forever without any maintenance.
Correct
Models degrade over time because real-world data changes (a phenomenon called 'data drift' or 'concept drift'). A model trained on 2023 shopping behaviour will be less accurate in 2025. Models need to be monitored, retrained, and updated periodically to maintain performance.
This mistake is common because people think of software as static (install once, use forever). ML models are statistical and depend on the environment; when the environment changes, the model's assumptions become invalid. AWS emphasises MLOps and model monitoring in the exam.
Mistake
Machine learning and deep learning both require huge datasets to work at all.
Correct
Traditional ML can work well with relatively small datasets (hundreds to thousands of rows). Deep learning generally requires very large datasets (hundreds of thousands to millions) to be effective. However, transfer learning and pre-trained models on AWS (like Amazon Rekognition) allow you to use DL with less data by starting from an existing model.
The 'big data' hype leads people to believe all ML needs massive data. The exam distinguishes between ML and DL partially based on data requirements, and transfer learning is a key concept to know.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
AI is the whole field of smart computers. ML is a way to make computers smart by letting them learn from data. DL is a more complex way of ML that uses brain-like structures (neural networks) to learn from huge amounts of data.
Yes. A thermostat that turns on the heat when the temperature drops below 18 degrees Celsius is AI (it performs a task that normally requires human decision-making) but it is not ML because it uses a fixed rule, not learned patterns from data.
For the AIF-C01 exam, you need to understand the concepts, not code. However, using AWS services like SageMaker typically requires some scripting (Python). But AWS also offers no-code services like Amazon Rekognition and Amazon Comprehend that you can call via API without writing ML code.
No. Deep learning requires massive datasets and powerful hardware. For simple problems with small datasets, traditional ML (like linear regression or decision trees) is often faster, cheaper, and easier to interpret. Deep learning excels at complex tasks like image/speech recognition.
The typical order is: 1) Collect data, 2) Clean and prepare data, 3) Choose an algorithm, 4) Train the model, 5) Evaluate the model, 6) Deploy the model, 7) Monitor the model. This is called the ML lifecycle.
Amazon SageMaker is for ML and DL. Amazon Rekognition (image/video analysis) and Amazon Polly (text-to-speech) are pre-built DL services. Amazon Lex (chatbots) is AI (can use ML). Simple rule-based decision systems might just use AWS Lambda with if-then rules (AI, not ML).
You've finished Introduction to AI and ML on AWS. Continue through the AIF-C01 study guide to build a complete picture of the exam.
Done with this chapter?