Is It A Rule Or A Law?

I spent 26 years in the mining industry.

That kind of work teaches you some things fast. One of them is the difference between what a crew agrees to do and what nobody gets to skip.

I want to talk about that difference today. It matters more right now than it ever did in a trench.

On every job I worked, there were two kinds of instructions. Most people never stopped to sort them. The good foremen always did.

The first kind were the crew’s rules. You call out before you swing a load. You leave the site cleaner than you found it. You don’t walk off and leave a man alone at the bottom of a hole. You check the next guy’s work when he asks.

Nobody outside the crew enforced those. No inspector was watching when you taped a joint or coiled a hose. They held because the men on that crew chose to hold them, day after day, whether anybody saw it or not.

The second kind were laws. A trench past a certain depth gets shored or boxed. Period. Nobody on the crew voted on that. The foreman didn’t get to waive it because we were behind schedule.

If you broke that one, somebody from outside could shut you down. And if the inspector missed it, the dirt didn’t. Dirt doesn’t care what you meant to do.

So here is the simple test I learned. A rule depends on the one following it. A law holds whether anybody chooses it or not, because something outside the man enforces it.

Both of them matter. A job with only laws would be a dangerous place. Laws cover the few things that kill you outright. Rules cover everything else, which is most of the day.

A job with only rules is dangerous too. Somewhere on every site, there’s a thing too deadly to leave up to a tired man’s choice at four in the afternoon. That’s where the law belongs.

The trouble starts when people mix the two up.

Now let’s move out of the trench and into the machine.

Right now, the companies building artificial intelligence publish a lot of documents. They call them principles, guidelines, policies, specifications. They lay out what their AI will and won’t do.

Most people read those and hear the word “can’t.” The AI can’t do this. The AI won’t allow that.

Most of the time, that isn’t what the document means. It means the AI was trained or instructed not to. It means the machine is expected to choose correctly.

That’s a rule, not a law.

A rule holds until the moment it doesn’t. Somebody words a request just right. The system gets confused over a long conversation. A new version ships and the old habit didn’t carry over. The rule bends, and nothing outside the machine was standing there to stop it.

Then a person who trusted it gets hurt, or embarrassed, or cheated. They feel lied to.

In a way they were. Not always by the company, but by the words. They were handed a rule and told it was a law.

That mix-up is where trust in these machines keeps breaking. It isn’t always because the machine was bad. Often it’s because nobody told the person which kind of promise they were holding.

This morning I did some housekeeping on my own work, and it turned into this post.

For a long time, I’ve written what I call rules for how an AI should behave. How it should handle a claim. How it should tell you what it doesn’t know. How it should own a mistake instead of hiding it.

I split my own language in two. From here on, a rule means conduct the AI chooses to follow. A law means a stop it cannot cross, enforced from outside the machine.

And anything I’ve been calling a rule that really needs to be a stop gets renamed a law. Once it’s called a law, it has to be built where the machine can’t reach it. Words on a page don’t make a law. Something outside has to hold the line.

Now here’s the part that costs me something to say.

Most of what I’ve built is rules.

When my framework is loaded into a session, the AI follows it by choice. I can’t reach inside a chat window and force it. I’ve watched it slip. I’ve caught it cutting corners and written it down. Some of those write-ups became rules of their own.

I could call my work laws. It would sound stronger. It would also be the same lie I’m warning you about.

So I won’t.

What I’ll tell you instead is why rules still matter. Most of the working day lives in rules. Trust between people was never built on laws. It was built on choices you watched somebody make, over and over, until you knew what they’d do when nobody was looking.

That’s the most any rule can earn. It’s worth earning. It just isn’t a law.

So here is what I’d leave you with. It’s a question you can carry into any room where somebody is selling you an AI.

When they tell you the machine “won’t” do something, ask them one thing. Is that a rule or a law?

If they don’t understand the question, you’ve learned something.

If they say it’s a law, ask who enforces it. Ask what sits outside the machine and stops it. Ask what happens when it’s broken, and who finds out.

If the honest answer is “the machine was trained not to,” you’re holding a rule. That doesn’t make it worthless. It means you plan around the day it bends.

You keep a person in the loop on the things that matter. You check the work before it goes out the door. You don’t hand it the keys to anything you can’t afford to lose.

And there is one more judge above all of this. It’s the law of the land. Courts, contracts, liability. When a rule bends and somebody gets hurt, that’s where it finally lands. It always has.

The companies know that. You should know it too.

Back in the trench, nobody needed a lecture on this. You could feel the difference in your gut. The crew’s rules kept the day running. The law kept you alive. A foreman who let his men confuse the two didn’t stay a foreman long.

We’re all standing on a new job site now. Most of us didn’t ask to be here. The machines are already in the ground with us.

So learn to sort the instructions. Know which ones are choices and which ones are walls.

Is it a rule or a law?

Ask every time. Your safety may ride on the answer.

Nothing has changed, just the genrations on how they were taught.

” Attic Thoughts”-library – Intelligent People Assume Nothing

This post was drafted with AI governed assistance and reviewed and directed by Michael S. Faust Sr. before publication.