Three words get used as though they meant the same thing, and the difference matters as soon as somebody asks you what the system will do.
Artificial intelligence is the broad field. Any technique that makes a machine behave in a way we would call intelligent belongs to it, including approaches with no learning in them at all. A rules engine that reads a written credit policy and decides eligibility is artificial intelligence.
Machine learning is the subset where behaviour is learned from data rather than written down. Nobody states the rule. The rule is inferred from examples, which is what makes it useful for problems too messy to specify and what makes it hard to explain afterwards.
Deep learning is the subset of machine learning that uses neural networks with many layers. The advantage is that it learns its own features. Classical machine learning needs somebody to decide that the useful signal in a photograph is edge density; a deep network works that out for itself, given enough examples. That is why deep learning took over computer vision, speech recognition and natural language processing, and why it needs far more data and compute than the alternatives.
Algorithm, model, training, inference
An algorithm is the procedure that learns. A model is the artefact that procedure produced. Training is running the algorithm over data to get the model. Inference is using the finished model to produce an answer for input it has not seen.
The distinction between training and inference decides most of your costs and most of your architecture. Training is expensive, occasional and done offline. Inference is cheap per call, constant, and sits in the path of a user waiting for a response.
Bias, fairness and fit
Three more terms that recur throughout the exam and mean something specific.
Bias has two meanings that are easy to confuse. Statistically, bias is error from a model too simple to capture the pattern. Socially, bias is a system producing systematically different outcomes for different groups. Both appear in the syllabus and the second is what governance is about.
Fairness is whether those different outcomes are justifiable. It is not a single measurement, and the definitions of it conflict with each other mathematically, which is why it stays a judgement rather than a threshold.
Fit describes how well a model matches its training data. A model that has learned the noise as well as the signal is overfitted and will do badly on anything new. A model too simple to capture the pattern is underfitted and does badly everywhere.
Where AWS puts each of these
The exam is vendor specific, so the vocabulary comes attached to services. Amazon SageMaker is the platform for building and training your own models. Amazon Bedrock serves foundation models you did not train. Amazon Rekognition, Amazon Transcribe, Amazon Comprehend, Amazon Polly, Amazon Translate and Amazon Lex are prebuilt services for vision, speech, text, and conversation, where you supply no training data at all.
That last group matters more than it looks. A large part of the exam is recognising when a managed service already solves the problem and training something yourself would be waste.