A Christmas AI Sweater Story

I was making an ugly Christmas sweater with AI. This should have been easy. In fact, it was going surprisingly well. I had an idea, gave it some reference images, and we worked through the design. I changed things. It changed things. We eventually got to a design I really liked.

Then I needed the artwork prepared for the company that was going to make the sweater. I said I needed four separate images:

Front.

Back.

Left sleeve.

Right sleeve.

Four images. Just four images and somehow, this became the most difficult part of the entire project.

AI made a picture with all four pieces together.

I said, no. I need them separately.

It made another picture with all four pieces together.

I explained again.

Another picture. All four pieces together.

At this point I started questioning my own ability to communicate in the English language.

So I got more specific.

FOUR. SEPARATE. IMAGES.

And I got… another image showing the four pieces together.

That was when I lost my composure and used language I generally don’t use when speaking to a computer. WTF?

The funny thing is that the AI had already done the hard part.

It had created the design.

It had changed the colors.

It had changed the wording.

It had replaced one graphic with another.

It had fixed tiny details I pointed out.

It had taken a ridiculous idea and turned it into something that actually looked like a real sweater.

But it could not seem to understand that when I said four separate images, I did not mean one image containing four things.

And that’s the part about AI that I find fascinating.

We spend so much time talking about how intelligent these systems are. And they really are astonishingly capable.

But then you ask for something incredibly simple and watch it confidently do something else.

That’s not really an intelligence problem.

It’s a “did you actually listen to me?” problem.

Anyone who has managed people knows that these are not the same thing.

I don’t need an teammate who can produce something beautiful while ignoring the deliverable.

I need someone who understands what we are trying to accomplish.

That’s where humans still have a pretty significant advantage.

A human designer would have understood why I needed four files. They would have understood that the files were going into another system. They would have understood that putting everything on one page wasn’t helpful.

Or, if they weren’t sure, they would have asked.

AI tends to do something different.

It guesses.

And sometimes the guess is spectacular.

Sometimes it’s spectacularly wrong.

The really interesting part is that it doesn’t necessarily know the difference. It can tell you, with absolute confidence, that it has understood you.

Then it does exactly what you just told it not to do.

Again.

And again.

And again.

Until you’re sitting at your desk, arguing with an artificial intelligence about the definition of the word separate.

That’s when you begin to appreciate the value of a human being who can look at a situation and say:

“Oh. I see what you mean.”

So no, I don’t think AI is going to replace humans anytime soon.

It is going to change the way we work. It is going to do things that would have seemed impossible a few years ago. And I will continue to use it because, despite this particular episode, it is incredibly useful.

But I’m also going to keep checking its work.

Because apparently somebody has to make sure the four sweater pictures are actually four pictures.

And yes.

The fact that an AI assistant drove me to profanity while making a Christmas sweater may be the most convincing evidence I’ve seen yet that humans still have a job.