Skip to content
hokkien.me
Menu

Lesson 7 of 21

Serving 3 · The Chef With a Thousand Eyes

Many Cooks, One Wok

  • 3 min
  • Transcript

Narrated by AI Sing, a cheerful plate of Hokkien Mee who runs his own stall.

AI Sing alone at the wok in the Saturday rush, nose over the first fry of eggs and garlic, ladle in the steam, the helper far away at the counter (AI Sing, the course mascot)

Listen to this lesson

The lesson

12 plates, 1 wok, no panic

It is Saturday night at my stall. A table of 12 just ordered 12 plates. My whole team? One helper at the front. He takes orders, collects money, serves the noodles. He never touches the wok. Behind the wok, only me, 1 pair of hands, and every plate must be perfect. From the counter, it looks like 1 cook with 1 wok. From inside, it is many tiny cooks.

A stack of tiny questions

With my eyes, I watch the eggs hit the oil. I ask 1 tiny question: is the egg golden and crispy yet? With my ears: is the sizzle lively? With my nose: are the eggs and garlic fragrant? With my hand: does the stir feel right? No single check knows what Hokkien Mee is. The answers appear only when everything is stacked. Golden, fragrant eggs mean the base is right. Simmering in stock, the noodles slowly drink up the taste. The stack grows deeper, the questions bigger, until the last layer calls the dish: a perfect plate of Hokkien Mee.

That is a neural network. It is many small decision units, stacked in layers. Each layer reads some data, asks 1 tiny question and passes the answer up. No single unit understands Hokkien Mee. The whole stack does. When people say deep learning, deep only means many layers.

How did my stack learn? The same way I did: thousands of plates. After every half-eaten plate, I tweaked my recipe. More crispy egg. Less noodle. More prawn shells in the stock. A network trains on thousands of labelled plates. Plate after plate, the stack gets closer to the teacher. The teacher is simple: a beautiful plate of Hokkien Mee that a happy customer finishes completely.

Deep learning at work today

Deep learning is the engine under most AI in the news. A hospital scanner flags a shadow before the radiologist sits down. A weather service reads radar and calls rain 90 minutes ahead. Both run on the same mathematics and science. Thousands of tiny questions, stacked into 1 judgement.

Run the checks in your head

Your turn. The eggs hit the oil. Run your own stack in your head: eyes for the crispy egg, ears for the sizzle, nose for the garlic, hand for the stir. What does the top of the stack call? I give you three seconds.

Pause here for three seconds.

Show the answer

… Perfect Hokkien Mee. The answers stacked, layer on layer, into 1 call.

Read this lesson again