Let’s play a game called “guess my age.”

You don’t know me. We are seated on either side of a wall in a room, you can’t see me, but we can hear each other.

The game goes like this: I tell you certain facts about myself, and you then use the information to guess my age.

I say “I play the banjo, I love reading and browsing Facebook. I graduated from high school several years ago. I have a pet dog.”

You say, “you are 42 years old”.

I say, “Not quite, you are off by 12. I’m actually 30.”

“Oh,” you say. “I should probably reevaluate my beliefs about people who browse Facebook in 2022. They are likely not as old as I previously thought.”

I now leave the room, only to be replaced with another individual who continues this very peculiar game.

“I play the flute, I hate reading and Twitter is my preferred social media fix,” they say. “I graduated only recently and I don’t have any pets.”

“You are most certainly 21,” you proclaim.

“Close but not quite,” comes the reply, “I’m actually 24.”

This goes on for several hours, you keep going through people and guesses, updating your beliefs along the way, until you have a very good idea of what facts about people are useful in predicting their age. E.g., you might conclude that a persons ability (or lack thereof) to play a musical instrument has no bearing on their age.

This is basically how machine learning algorithms work.