Part 2 · How AI Actually Works
How AI Models Are Trained (The Hokkien Mee Analogy)
Picture a hawker at the wok, frying a plate of Hokkien Mee. Building an AI system runs on almost the same recipe. AI learns its patterns from examples, and the order of the work matters just as much as it does at the wok.
Why This Matters
Most people picture building AI as a single step: switch it on. It is not. Understand the real steps and you stop expecting instant results, and you start asking your technical teams the right questions. This is Understanding AI, pAIRI Dimension 1: the foundation the rest rests on. By the end you will be able to:
- Describe the seven steps of building an AI system using the Hokkien Mee analogy.
- Explain why preparing the data takes most of the effort.
- Map a real AI project to those same seven steps.
The Recipe Behind Every AI System
Cooking Hokkien Mee is a sequence of steps, and so is building an AI system. First, you decide what you are making. A hawker does not fire up the wok at random; you choose the dish. In AI, that means defining the problem you want to solve.
Then come the ingredients. You buy your prawns, noodles and eggs: you collect the data. And here is the part nobody photographs for the menu: preparing those ingredients. Washing, peeling, deveining, slicing. In real AI projects, this cleaning and preparing of data is where most of the effort in the project goes. If the ingredients are bad, the noodles will be bad, no matter how good the cook is.
- Decide the dish
- Buy the ingredients
- Prep the ingredients
- Pick the recipe
- Fire the wok
- Taste and adjust
- Serve it
Only now do you choose your recipe (the algorithm) and start cooking, which is training the model on your prepared data. Then you taste and adjust: too bland, add stock. That is evaluating the model and tuning it until it performs. Finally you serve it (you deploy the model to real users) and you keep watching how it lands, because tastes change. Seven steps. The wok is the smallest part; the preparation is the real work.
Key Takeaway
The wok is the smallest part; the preparation is the real work.
Check your understanding
Self-check only. Nothing is scored and nothing is recorded; open an answer to compare it with your own.
A retail company wants to build an AI system that predicts which products will sell out next month. They have three years of sales data but it contains thousands of missing entries, inconsistent product names, and duplicate rows. Using the Hokkien Mee analogy, which step should they focus on first?
- Step 5 (Fire the wok / Train the model), start training immediately and fix problems as they appear.
- Step 3 (Prepare the ingredients / Data cleaning), fix the missing entries, standardise product names, and remove duplicates before training.
- Step 1 (Decide what to cook / Define the problem), go back and redefine the problem since the data is messy.
- Step 2 (Buy the ingredients / Gather more data), collect additional clean data to replace the messy records.
Show the answer
Step 3 (Prepare the ingredients / Data cleaning), fix the missing entries, standardise product names, and remove duplicates before training.
An AI model for detecting insurance fraud was ninety-five per cent accurate when tested internally, but after deployment (Step 7), it started flagging legitimate claims as fraudulent far more often than expected. Using the Hokkien Mee analogy, which step addresses this?
- Step 5 (Fire the wok / Retrain), the algorithm needs a complete rebuild with different techniques.
- Step 6 (Taste and adjust / Re-evaluate), the initial testing was not rigorous enough and should be repeated.
- Step 7 (Serve and watch / Monitor and retrain), real-world patterns differ from the training data, so the model needs updating with new examples.
- Step 1 (Decide what to cook / Redefine the problem), fraud detection may be too complex for AI.
Show the answer
Step 7 (Serve and watch / Monitor and retrain), real-world patterns differ from the training data, so the model needs updating with new examples.
Saved in this browser only. Carry progress between devices with the resume link on the course page.