Fidgety Crayon

If I had to put it poetically, it's the difference between writing your thoughts in a notebook and releasing a parrot into a crowded dinner party, letting it listen to six hundred conversations at once, and then treating the resulting word salad as though it possesses intention.

The internet is a library.

A tool.

A search engine helps you locate information.

None of those things are attempting to generate the answer for you.

You still have to read.

You still have to evaluate.

You still have to decide whether a source is trustworthy, whether an argument makes sense, whether a claim can be substantiated.

You still have to do the thinking.

A large language model does something fundamentally different.

It takes an input and generates an output based on patterns found in its training data. It does not know whether the answer is true. It does not understand the argument it is making. It cannot distinguish between a correct statement and an incorrect one in the way a reasoning mind can.

What it does is predict what a convincing answer would probably look like.

Fact checking matters.

Source verification matters.

Understanding where information comes from matters.

Knowing how an argument was constructed matters.

Because if your process for writing becomes "generate paragraph, accept paragraph," then you've skipped the very thing writing is supposed to cultivate.

And that's the part that concerns me.

Not because I hate the concept of machine learning.

I don't.

What people commonly call AI is, at its core, a method of processing and generating information through pattern recognition. As a concept, it's morally neutral. It's also not particularly new.

The problem is that a lot of the public conversation around it treats these systems as though they are intelligent in the way people are intelligent, and many companies are more than happy to encourage that perception through the way they're marketed and sold.

But the systems currently available are not reasoning minds. They do not understand what they're saying. They do not possess intent, beliefs, or comprehension. They do not think through problems the way humans do. They generate outputs based on patterns found in data.

My issue isn't with the existence of the technology.

My issue is with the way large language models are currently marketed, presented, and trained.

They're sold as tools that can help you write, help you think, help you learn.

But those are three very different things.

Helping someone write should mean supporting the writing process.

Helping someone think should mean encouraging critical engagement.

Helping someone learn should mean deepening understanding.

What these systems often do instead is produce something that resembles those outcomes from the outside.

Without necessarily requiring the process that creates them.

Writing is not merely the act of producing words.

Writing is thinking made visible.

"AI", and by that I mean large language models using machine learning, are not performing independent thought. They are not reasoning in the way people do. They are taking the data they have been trained on, remixing patterns from that dataset, and generating a response that statistically resembles the kinds of answers that would fit the input.

THAT is the difference.

Writing the process of wrestling with an idea long enough to discover what you actually believe about it.

If you outsource that process entirely, you end up with text, but you've skipped the part where the writing was supposed to happen.

There's also the question of the source of the training data.

These systems are built on a staggering amount of stolen human work.

Entire lifetimes of creative and intellectual labor compressed into datasets, gathered without consent from the people who created them.

And even if someone doesn't find that ethically troubling, and I would argue there's very good reason to, it still creates an enormous tangle of copyright, ownership, attribution, and compensation issues that society is currently scrambling to understand in real time.

The technology arrived before the rules.

And we're all now living in the consequences of that.

In the original post I ended by saying:

"I spent so much of my life learning how to write. I shouldn't have to unlearn that because some computer algorithm learned from me."

I stand by that.

Because at the end of the day, my issue isn't that a machine can generate sentences.

It's that somewhere along the line we've started treating the generation of sentences as interchangeable with the act of having something to say.