Artificial intelligence, machine learning and deep learning are three nested ideas, not three competing technologies. This lesson defines each one, plus the terms the exam expects you to know alongside them: neural networks, computer vision, natural language processing, models, algorithms, training and inference.
The broad goal of AI
Artificial intelligence (AI) is the umbrella term for computer systems that perform tasks we normally associate with human thinking: recognising images, understanding speech, making decisions or holding a conversation. AI describes the goal, not a particular method. Early AI systems were often hand-written rules ("if the temperature is above X, raise an alert"). Rules work when the problem is small and stable, but they break down when there are too many cases to write by hand.
How machine learning learns from data
Machine learning (ML) is the part of AI where a system learns patterns from examples instead of following rules a person wrote. You show it historical data, such as past loan applications and whether each was repaid, and it works out the relationships itself. ML is the most common way AI is built today. Its strength is handling problems where the patterns are real but too complex to spell out, like spotting fraud among millions of card payments.
Deep learning and neural networks
Deep learning is a branch of ML that uses neural networks with many layers. A neural network is a structure loosely inspired by the brain: layers of connected nodes, where each layer passes a slightly more refined signal to the next. "Deep" refers to the number of layers. Deep learning shines with unstructured data such as images, audio and free text, because it learns which features matter on its own. Traditional ML usually needs people to choose those features by hand. The cost is that deep learning needs much more data and computing power.
Computer vision and natural language processing
Two application areas come up constantly. Computer vision is AI that interprets images and video: reading a licence plate, finding a defect on a production line, or counting people in a store. Natural language processing (NLP) is AI that works with human language: detecting the sentiment of a review, pulling names out of a contract, or translating text. Both are now mostly built with deep learning.
Models, algorithms, training and inference
These four words are easy to mix up:
| Term | What it is |
|---|---|
| Algorithm | The learning procedure, a recipe for finding patterns in data |
| Training | Running the algorithm over data to learn the patterns |
| Model | The result of training: the learned patterns, saved so they can be reused |
| Inference | Using a trained model to make a prediction on new data |
The algorithm is the method; the model is what the method produced.
On the exam
Expect questions that test the nesting: deep learning is inside ML, which is inside AI. If a scenario stresses images, audio or text with no hand-built features, deep learning is usually the intended answer. If an option says "the algorithm makes predictions in production", that's the trap: the model makes predictions, and doing so is called inference.