How Does Google Know That?" A Kitchen Chat with My Mom About AI

Characters:
Me: Your friendly guide to technology.
Mom: Curious, smart, and the master of her kitchen.
(Scene: We are sitting in the living room, Mom is scrolling on her phone.)
Me: Hey Mom, have you ever wondered how your phone just knows what you mean? When you search for something on Google, it gives you the perfect answer, almost like it can read your mind.
Mom: All the time! Just the other day, I was looking for a new recipe. I typed "healthy evening snack without frying," and it gave me so many great ideas. It's like magic. How does it know "healthy" means "not fried"?
Me: Well, it's not magic, but it's very clever. It’s all because of something called "vector embeddings."
Mom: Vector what? That sounds complicated. Is it some kind of code?
Me: It is, but you already understand the main idea perfectly. Think about your kitchen. It's the most organized place in the house, right?
Mom: Of course! If it wasn't, I'd never find anything.
Me: Exactly! Now, imagine a computer's brain before it learns. It sees all the words in the world like a giant, messy pile on the floor. The words "spoon," "salt," "car," and "dish soap" are all mixed up. The computer knows these words exist, but it has no idea what they mean or that a "spoon" is more related to a "fork" than it is to a "car."
Mom: What a mess! That sounds useless. You could never cook in a kitchen like that.
Me: Precisely! So, computer scientists decided to teach the AI to organize that mess, just like you organize your kitchen. That organizing process is what "vector embeddings" are all about.
Mom: Ah, so you mean how I keep all my spices in the masala dabba? And all the different daals are on one shelf, next to the rice?
Me: Yes, that's it exactly! In the AI's "kitchen," every word gets a specific location based on its meaning. All the fruits like apple
, banana
, and mango
are in one "fruit basket." All the utensils like spoon
, fork
, and plate
are in one "drawer." And the dish soap
is under the sink, far away from the food. Each location has an address a set of coordinates and that address is the vector.
Mom: Okay, I get it. So things with similar meanings are neighbors, and different things live far apart. Simple enough.
Me: But here's the truly magical part. The AI can also understand relationships. It can learn directions within the kitchen.
Mom: Directions? What do you mean? Like a GPS?
Me: Pretty much! For example, scientists discovered you can do math with words. If you take the word King
, subtract Man
, and add Woman
, the AI knows the answer is Queen
.
King−Man+Woman≈Queen
Mom: Wait, you can do math with words? That's amazing!
Me: You can! Let's try it with your kitchen. If you start at Pot
, and you tell the AI to subtract the idea of using it on a Stove
, but add the idea of using it in an Oven
... where do you think it would lead?
Mom: Hmm, a pot for the oven... it would have to be a Baking Dish
!
Me: You got it! And that's how it all works. The AI understands the function and relationship between things, not just what they are.
Mom: So that's how it knew what I meant by "healthy evening snack"! It knows "healthy" is close to "baked" or "steamed," but far away from "deep-fried."
Me: Exactly! This is the secret behind Google, the movie recommendations on Netflix, and even those chatbots that can have entire conversations with you. They all rely on this super-organized "word kitchen."
Mom: Wow. So you're saying all this fancy technology is based on the same logic I use to keep my kitchen tidy?
Me: Yep! You've been a vector embedding expert all along and didn't even know it.
Mom: (Laughs) I'll have to remember that. Now, speaking of snacks, are you hungry?
Subscribe to my newsletter
Read articles from Mohit Chamolla directly inside your inbox. Subscribe to the newsletter, and don't miss out.
Written by
