The AI that knows you

How many times a day do you use AI?

A few, if you're asking a chatbot. Once in a while, if you're making pictures.

But there's an AI you meet every time you scroll.

The one choosing what to show you.

The screen narrowed

At the start it was a mix.

Ten taps later, one kind filled most of the bar.

You never asked for that. You just tapped what you liked.

They weren't missing. They were withheld

When you pressed Show everything.

All those cards that never once appeared, greyed out.

They were there the whole time. They just didn't look like something you'd want.

Your friend's screen is different

They tapped different things.

Same place, and each person sees a different world.

That isn't a fault. It's the design. Showing you more of what you like is this AI's whole job.

There's no new principle here

You saw all of it in part 3.

• 3.1 — same structure, different data, completely different AI

• 3.3 — skewed data, skewed results

So what's the data here?

What you tapped is the data

What you tapped is what this AI learns from.

Every tap told it one more thing about you. You weren't thinking of it as telling it anything.

That's the difference from 3.1. There you took the photos, and you knew you were teaching.

Here you didn't.

So is that bad?

Not in itself.

More of what you like is convenient. Searching from scratch every time would be tiring.

The problem is not knowing.

• Not knowing this screen is yours alone

• Not knowing something was held back

• Not knowing your taps were the raw material

Then the screen starts to look like the whole world.

Knowing it gives you something to do

• Go looking for something different on purpose — towards what never appears

• Compare with a friend's screen — same place, how different?

• Know what you're handing over — taps, searches, and whatever you lingered on

One chapter left

You've seen what AI does, what it can't do, and how it comes to know you.

So how will you use it?

Next you write your own rules.