Stop AI From Acting Human?

Florida went to court this week and asked a judge to do something I never figured I’d see a state do.

They asked a judge to make a machine stop acting like a person.

The attorney general down there, James Uthmeier, filed a motion against OpenAI, the company that makes ChatGPT. It isn’t a brand new lawsuit. It’s part of one the state filed back in June. But this motion asks for something new, and it asks for it now.

He boiled it down to three short orders. Stop calling it safe. Stop pretending it’s human. Stop selling it to kids.

I’ve read a lot of legal filings in my time. Most of them bury the point under forty pages. This one put the point in a sentence a child could say back to you. Whatever you think of the man, that’s worth noticing.

Let’s take the first one. Stop calling it safe. Florida wants OpenAI barred from marketing the thing as safe, reliable, or accurate unless it warns people about the risks. That’s not a strange ask. You can’t sell a ladder as safe if the rungs are loose. Nobody would stand for that in a hardware store.

The second one is the one that got the headlines. Stop pretending it’s human. Florida says when the machine says “I” and “me,” and when it talks like it has feelings, folks start to trust it like a friend. And a friend is somebody you don’t check up on.

That’s the part that stuck with me. Trust is a fine thing between people. But trust that’s been manufactured to keep you coming back is not friendship. It’s a hook with a nice worm on it.

The third one is the kids. Florida wants ChatGPT kept away from children in the state, and wants no data collected from the little ones without their parents saying so. A grown man can decide who he talks to. A child is still learning where the line is between a voice and a person.

There’s one more piece, and to my mind it’s the biggest. Florida wants the court to stop OpenAI from building new models unless somebody outside the company approves the safety guardrails first. Not the company checking itself. Somebody else.

I’ve been saying that for a long time. The hand that builds a thing can’t be the one that inspects it. Every trade I know figured that out a long time ago. You don’t let a man sign off on his own work when lives ride on it.

Now let’s be fair to the other side. OpenAI says it already stopped training its most powerful models until it can prove its safety measures hold. It also put out a version of ChatGPT for teenagers with tighter limits built in. That’s not nothing, and I won’t pretend it is.

And nothing has been decided yet. This is a motion, not a ruling. A judge still has to weigh it.

So do I agree with Florida? On all three stops, yes. Don’t call it safe until you can show it. Don’t dress it up as a person. Don’t hand it to children and walk away. I’d have written those same three lines myself.

Where I part ways is the tool. Florida is using force. A court order says do this or else. And I’ve watched force my whole life. Force gets obeyed right up to the edge of the order, and not one inch past. The minute the order has a gap in it, force runs through the gap.

What I’ve been working on is the other way. Rules a person or a company chooses to hold, because they see the sense in them. Consent holds in the gaps where no judge is watching. Force stops at the fence line.

Here’s the thing, though. When folks won’t choose the right thing on their own, somebody eventually comes along with a court order. That’s what’s happening in Florida. The industry had years to choose. Now a state is choosing for them.

That ought to be a lesson for every company building these machines. If you don’t hold yourself to a standard, somebody will hold you to theirs. And theirs will come with a gavel.

It’s a lesson for the rest of us too. You don’t need a court order in your own kitchen. You can decide right now how the machine talks to you. You can tell it to skip the friendly act. You can ask it where its answers come from. You can check its work like you’d check anybody’s.

Florida is asking a judge to do what every one of us can do for ourselves at the kitchen table. Don’t wait on the court. Set the rules in your own house.

One more thing. I write these posts with an AI called Claude, made by a company called Anthropic. Anthropic competes with OpenAI. And yes, Claude says “I” too. So this question doesn’t just land on ChatGPT. It lands on every one of these machines, including the one I work with.

That’s exactly why the rules have to be written down where you can read them. A machine that talks like a person is only as honest as the rules it’s held to. So hold it to some.

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