Your own rules for AI
You've come a long way.
From photographs being numbers, through learning as fixing-by-the-amount-you're-wrong, guessing the next piece, and making pictures out of mess.
One thing left. How will you use it?
Remember what you wrote at the start?
In 1.3 you wrote something on a blank card. What AI would be doing ten years from now.
That was right at the beginning. What does it look like now?
Three tests for making your rules
The tests piled up as you went. Here they are together.
One — can I check the result?
The test from 4.5.
Code you can run. Writing you can read. But the plot of a book means digging through a library, and what happened in class today can't be checked at all.
Handing over what you can't check, and then believing it, is the dangerous one.
Two — is there someone answerable when it goes wrong?
The test from 6.1.
You can't take it up with an AI. You can't punish it. Decide who can put it right before you start.
Three — is this something I get better at by doing?
This one is new.
Calculators exist and you still learn multiplication. Why?
Because some things only grow by doing them. Writing. Working a problem through. Trying again after getting it wrong.
Let an AI do it and you still get the result. But you don't get stronger.
So it might split like this
| Examples | |
|---|---|
| Fine to hand over | summarising long text, tidying sentences, first drafts, running code |
| Be careful | finding facts, anything you'll send to someone, judging people |
| You grow by doing | having the first idea, explaining why, spotting what's wrong |
That's where this guide landed. Yours may be different.
Now write your three lines
Go back up to the lab and write your three rules.
What you'll hand over
What you'll do yourself, because that's how you grow
What you'll always check first
Rules handed to you rarely stick. That's why you write these.
When you're done, press Take it as a file.
What happens next?
You saw it in 1.3. People's predictions missed, again and again. Things declared impossible got done; things declared imminent took decades.
So what you write here may be wrong too. Still, from everything you've seen, this much holds.
What AI will get better at — doing what it already does, faster and more accurately. Words, pictures, sound, code: things where being good at it can be measured.
What people will keep doing — deciding what to do, saying why it matters, being answerable when it goes wrong.
The first is making answers. The second is setting the question.
AI is good at answers. What to ask is still ours.
One last thing
This guide wasn't written to make you wary of AI.
It was written so you'd know what it is before you use it.
Not knowing, it looks remarkable. Knowing, it looks like a tool. Tools aren't frightening. They just have to be understood to be used well.
Anyone who got this far understands.