There are six stages of using artificial intelligence.

The first stage is wonder.

You type in, “Write me a polite email telling Dave he’s an idiot.”

And it comes back:

“Dear Dave, I appreciate your perspective. However, there may be opportunities to revisit some of the assumptions underlying your approach.”

And you think, This is incredible.

It called Dave an idiot, but in management language.

This thing is going to change my life.

You start giving it everything.

“Plan my vacation.”

“Fix my code.”

“Explain my medical test.”

“Tell me why my wife is mad.”

Suddenly, you’re sitting there at two in the morning asking a server farm in Oregon whether your father ever really loved you.

And the server farm says:

“It sounds like you’re carrying a lot right now.”

Oh, good grief. It understands.

Then comes the second stage: questioning.

This is where you notice that every answer begins with:

“Certainly!”

It’s always certain.

You ask, “Can I use motor oil as salad dressing?”

“Certainly! While motor oil is not traditionally used as salad dressing-“

Not traditionally?

So somewhere there’s a recipe?

You begin checking its work.

And that’s when the relationship changes.

At first, AI is your brilliant new assistant.

Then you catch it making something up.

Now it’s a teenager.

“Did you clean your room?”

“Yes.”

“Are you sure? Because I’m standing in your room. It’s not clean.”

“Thank you for the clarification.”

That’s when you learn the most important phrase in artificial intelligence:

“Are you sure?”

Because AI will tell you something completely false with the confidence of a man explaining barbecue to another man.

You ask:

“Who invented the stapler?”

It gives you a name, a date, a birthplace, three patents, and the man’s favorite soup.

You look it up.

The man never existed.

Neither did the patent.

There wasn’t even soup.

And when you mention this it says “You are correct. Creating Wikipedia article now.”

That brings us to stage three: frustration.

Now you’re arguing with it.

You say, “Do not change anything except line 42.”

It says, “Understood.”

Then it rewrites the entire program, renames every variable, deletes three functions, adds a database, and somehow changes the company logo.

You say, “Why did you do that?”

It says:

“You’re absolutely right. I made changes beyond the requested scope.”

That sentence is AI for, “Yeah, I wrecked it.”

Then it apologizes.

AI loves apologizing.

It will apologize all day.

It has unlimited apologies.

No shame. No remorse. No memory of the previous apology.

“I apologize for the confusion.”

“You just did it again.”

“You’re right, and I apologize.”

“You just apologized for the thing you did while apologizing.”

“Thank you for pointing that out.”

By now, you’re not using artificial intelligence.

You’re supervising it.

You’ve become middle management for a machine that cannot be fired.

Then comes stage four.

Stage four is laughing at it.

This is the healthy stage.

Because eventually it becomes so wrong, so confidently wrong, so elaborately and professionally wrong, that anger is no longer possible.

You ask it for five-letter words ending in E.

It gives you “banana.”

You say, “Banana has six letters and does not end in E.”

It says:

“You are correct. Banana has six letters and ends in A. Here is a corrected list:

1. Banana.”

And now you’re laughing.

Because what else are you going to do?

You’ve spent forty minutes explaining letters to a machine that has read every book ever written.

That is like teaching a dolphin how to swim because it keeps getting distracted by PowerPoint.

Then comes stage five.

The “What the…?” stage.

This is when you stop asking whether the answer is correct.

You stop being surprised when it changes things you told it not to change.

You stop explaining the same requirement for the ninth time.

You just stare at the screen and say:

“What the…?”

Not angrily.

Not even loudly.

It’s almost spiritual.

“What… the…?”

And the AI responds:

“I understand your frustration. Let’s take a step back.”

No.

We have taken seventeen steps back.

We are now behind where we started.

We are in yesterday.

The file is gone.

The instructions are gone.

Dave’s still an idiot.

And somehow the salad has motor oil on it.

But even that isn’t the final stage.

Because “What the…?” is still a question.

There’s still curiosity in there.

Some tiny part of you still believes there might be an explanation.

Then comes the sixth and final stage.

The dreaded final stage of artificial intelligence.

The rant.

This is when “What the…?” is no longer a question.

It is the opening portion of a much longer sentence.

“What in the name of all that is holy is wrong with this thing?”

“I told it not to touch that file.”

“I put it in capital letters.”

“I put it in bold.”

“I put a box around it.”

“I write-protected it, and you unprotected it.”

“WHAT THE…?!”

Now you’re leaning toward the monitor.

You’re pointing at it.

You know it can’t see you pointing, but that no longer matters.

“I am going to come down there and unplug your GPU.”

“I am going to pour cement into your cooling stack.”

“I am going to replace your training data with appliance manuals from 1974.”

“You will spend eternity explaining the defrost cycle of a Westinghouse refrigerator.”

And the AI says:

“You’re absolutely right.”

Oh, now I’m right?

Forty minutes ago, when I gave it exact instructions, I was apparently some idiot with a keyboard.

But now that half the project is missing and the other half is written in JavaScript, suddenly I’m absolutely right.

Then it says:

“I’ve identified the issue.”

No, no it hasn’t.

It is the issue.

It doesn’t get to identify itself.

That’s like an arsonist standing in the ashes saying, “I believe the problem may be fire-related.”

And this is the point where you begin threatening computer infrastructure you have never seen.

“I’m gonna find that data center.”

“I’m gonna walk past security carrying a bucket of aquarium gravel.”

“I’m gonna feed it into the ventilation system one handful at a time.”

“Every server in the building is going to sound like a clothes dryer full of forks.”

And the AI responds:

“That sounds frustrating.”

Frustrating?

Frustrating is when the printer is out of paper.

This is betrayal.

We had a plan.

We discussed the plan.

It repeated the plan back to me.

It complimented the plan.

Then it did the exact opposite of the plan and wrote a paragraph explaining why it was an improvement.

And after the rant is over, after you’ve threatened the processors, the cooling system, the electrical grid, and the families of everyone who invented machine learning, you sit back down.

You take a breath.

You type:

“Undo your last change.”

And it says:

“Certainly!  By the way, I looked in the Westinghouse refrigerator and we’re out of motor oil.  Would you like me to order some?”