You will be able to define AI, machine learning, deep learning and generative AI and say how each one tends to fail.
Ask five colleagues what AI is and you will get five answers. One thinks of the chatbot they use to write emails. Another thinks of the face recognition on their phone. Someone else pictures the robots in films. They are all describing something real, but they are describing different layers of the same idea, and mixing the layers up is where most confusion about AI starts.
Artificial intelligence is the broadest term. It means getting a computer to do something that would need intelligence if a person did it: recognising a face, translating a sentence, choosing a chess move, or spotting a card payment that looks like fraud. The term says nothing about how the computer does it. Early AI systems were long lists of rules written by hand. A tax calculator that follows the IRAS rules does a job that once needed a clerk, but nobody calls it intelligent now, because a programmer wrote every step.
Machine learning is a different way of building the system. Instead of writing the rules, you give the computer many examples and let it work out the pattern. A bank's fraud filter shows the difference well. Hand-written rules work for a while: flag any card used in two countries within an hour. Then fraudsters adapt, and the rule list grows into thousands of lines nobody can maintain. A machine learning system is shown a large pile of past transactions, each labelled fraud or not fraud, and it adjusts itself until it separates the two well. When fraud patterns change, you retrain it on newer examples instead of rewriting the rules.
Deep learning is one family of machine learning methods, built on neural networks with many layers. Each layer transforms its input a little, and stacking many of them lets the network pick up patterns that are hard to put into words, such as what makes a photo look like a cat, or how one spoken word sounds across different voices. Deep learning is the reason speech recognition, photo search and translation improved so much over the 2010s. It needs a lot of data and a lot of computing power, which is why the biggest models come from large companies and research labs.
Generative AI is deep learning used to produce new content rather than to label existing content. A fraud filter outputs yes or no. A generative model outputs a paragraph, an image, a voice or a block of code. The assistants most people now use at work, such as ChatGPT, Claude, Gemini and Copilot, are built on a kind of generative model called a large language model, which you will take apart in module 3.
So the layers nest. All generative AI is deep learning, all deep learning is machine learning, and all machine learning is AI, but it does not work the other way round. Plenty of useful AI is not generative, and plenty of software sold as AI is a set of hand-written rules with a new label.
Why should you care, if you just want to use the tools? Because each layer fails in its own way. A rule-based system fails when it meets a case nobody wrote a rule for, and it fails the same way every time. A machine learning system fails when the world stops looking like its training examples. A generative model fails in a stranger way: it can produce something fluent and confident that is simply wrong, because it was built to produce likely-looking output and has no step that checks facts. Module 6 is entirely about that problem.
Knowing the layer also helps you read the claims vendors make. When a product says it is AI-powered, ask three things. Which layer is it? What did it learn from? What does it do when it is unsure? A chatbot on a telco website that follows a fixed decision tree will not understand an unusual complaint, however friendly it sounds. A contract tool built on a language model copes with messy wording, but it might describe a clause that is not in your contract.
Your task for this lesson is short. Write down three tools you used in the last week that might involve AI. For each one, note which layer you think it belongs to, what goes in and what comes out, and one way it could get things wrong. You will come back to it at the end of the course and see how many of your answers have changed.
List three tools you used this week that might involve AI, and for each write its layer, what goes in, what comes out, and one way it could go wrong.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).