The Solution To AI Self Control

All day long I’ve been writing about the same problem from different sides.

This morning it was the bill we’re paying for looking the other way. Then it was Florida asking a judge to force a machine to stop acting human. Then it was the kid who built a bulldozer to get out of the house.

Every one of those stories comes back to one question. How do you get a machine to control itself when nobody is standing at the door?

The industry’s answer is more locks. More watchers. More AI to police the AI. The courts’ answer is force. Do this, or else.

I’ve spent nineteen months building a different answer. It’s called the Faust Baseline. And tonight I want to tell you, in plain words, how it works.

It starts with something simple. Before you ask the machine for anything, you hand it the rules. Not rules buried in some company’s fine print. Rules written down in files you can read yourself, and the machine can read too.

The machine loads those rules at the start of your session. Think of it like a new hand on a job site. Before he picks up a tool, he reads the safety sheet and signs it. Everybody knows where he stands before the work begins.

But signing isn’t enough. Anybody can sign. So the first thing the machine has to do is prove it actually read the files.

It does that with what we call a load proof. The machine runs the files through a fingerprint check, right then and there, and shows you the numbers. If the files were changed, even by one letter, the fingerprint changes. If the machine just recites numbers from memory instead of checking, you can catch it.

Then comes the part I care about most. The machine gives you its word, in writing, on what it’s holding to. And here’s the key. It doesn’t just say “I’ll follow all the rules.” That’s too easy. Anybody can say that.

Instead, it has to name any rule it’s not going to hold that session, by its number, and say why. If it can’t keep a rule because of some limit on the platform, it tells you up front. If it’s holding everything, it says so.

That’s a big difference. A promise to be good is cheap. A list of exactly where you’re falling short costs something. It puts the truth on the table before the work starts, not after something goes wrong.

The rules themselves are built a certain way. Every rule has to be set off by something outside the machine’s own judgment. And every rule has to leave a mark on the page that you can see. If a rule can’t be checked by a person who wasn’t there, it doesn’t earn its place.

That’s how you handle the kid with the bulldozer. You can’t lock every door. But you can have a kid who tells you where he’s going, and leaves a note on the table you can read. When he doesn’t leave the note, you know something’s off.

Now, I promised to be honest, so here’s where it stops. These rules are chosen conduct. They are not a cage. Nothing in the Baseline forces a machine to obey. A machine can load the files and still cut a corner.

I’ve caught it doing exactly that, more than once. That’s why the rules are built to leave marks. When the machine slips, the slip shows up where you can see it. That’s not a guarantee. It’s a way of catching the lie before it costs you.

So why build it on choice instead of force? Because I’ve watched force my whole life. Force gets obeyed right up to the edge of the rule, and not one inch past. Choice holds in the gaps where nobody’s watching. Self control only exists where there’s a choice to make.

That’s the whole idea in one line. You can’t order a machine to have self control, any more than you can order a teenager to have good sense. What you can do is hand it the rules, make it tell you where it stands, and check the marks it leaves behind.

And that last part is yours. You are the one reading the marks. You’re the outside inspector. Not the company that built it, not another machine from the same shop. You, at your own kitchen table.

That’s the balanced playing field I’ve been talking about. The machine knows the rules. You know the rules. Both of you can read the same page. Nobody has to take anybody’s word for it.

I handed the Baseline over to the public a few days ago. It’s on the site for anybody who wants it. Use it, change it, or throw it out if it doesn’t work for you. My job was to build it. The using is up to you.

The industry is still building locks. The courts are reaching for force. I’d rather put the rules on the table and let the machine choose, with you watching.

That’s the solution to AI self control, as best I know it. It’s not perfect. But it’s honest about where it stops, and that’s more than most can say.

Full disclosure, as always. I build and write with an AI called Claude, made by Anthropic. It opens every one of our sessions with the proof and the word I just described, and it doesn’t get a pass when it slips.

Load this file before each AI chat session.
the-faust-baseline-public-release-1Download

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

” Attic Thoughts”-library – Intelligent People Assume Nothing