Who Wrote This?

I understand why people are uneasy about AI-assisted writing. There is a real difference between writing something you have actually thought through and generating something that only sounds like thought. Readers are right to care about that distinction.

What bothers me is how quickly the conversation gets flattened. “AI-generated” becomes a full explanation. A moral category. A verdict. The surface gets treated as the truth.

That is what feels wrong about Substack’s scanner. It looks at language patterns and offers a score, but that score cannot tell you who did the thinking, who exercised judgment, or who remains responsible for the work. It measures traces in prose. It does not measure authorship.

This is where the distinctions matter.

Using AI to replace thought in writing or any other matter is wrong. Asking a system to produce something you have not meaningfully considered, judged, or earned is not authorship. It is outsourcing the work that actually matters.

But that is not the only way these tools are used.

Sometimes AI helps communicate existing thought. It helps clarify structure, reduce noise, tighten language, and make an idea more legible. That is not the absence of authorship. It is assisted expression.

Sometimes AI participates in the thinking process itself. It helps question assumptions, compare framings, surface connections, and pressure-test a half-formed idea until the writer better understands what they actually believe. That is not the same thing as having the machine do the thinking for you. It can be part of thinking.

Those differences are not minor. They are the whole issue.

My own writing practice lives in the second and third categories.

I do not use AI because I have nothing to say. I use it because I often have too much to say at once. My thinking is associative, layered, and nonlinear. I can feel the pattern before I can always explain it clearly. The difficulty is not generating material. The difficulty is translating it into clean, sequential language without losing the signal.

AI helps with that. It helps me cut away noise, test structure, and find the clearest path through an idea. The conversation often reveals connections I sensed but hadn't yet found the language for.

But it does not decide what is true.

It does not decide what matters, what sounds like me, what should be removed, or what the final piece is actually saying. That judgment remains mine. The work is still mine to stand behind.

That’s why I don’t like the scanner.

It pretends to clarify the issue while actually obscuring it. It invites people to mistake a surface reaction for an understanding of process. It encourages the lazy conclusion that if AI was involved, the work must therefore be less human, thoughtful, or earned.

It mistakes classification for understanding.

And this is part of a broader cultural pattern. People increasingly form strong opinions from headlines, reaction clips, summaries, and labels without looking underneath them. They want the category before they understand the process. “AI-generated” becomes one more shortcut.

The scanner fits neatly into that impulse. It offers something that feels like evidence while avoiding the harder question.

The harder question is not whether AI touched the sentence.

The harder question is who owned the meaning.

Who brought the experience? Who recognized the signal? Who made the selections? Who rejected what sounded polished but false? Who remained responsible for the final argument?

That is what authorship is starting to mean to me.

Writing has never been completely solitary. People think with editors, notebooks, conversations, drafts, and reference materials. AI is not identical to those things, and I don’t think we should pretend it is. But neither should we pretend that any tool involvement automatically erases human ownership.

The line is not “machine touched text, therefore fraud.”

The line is whether judgment stayed human.

There are real risks here. AI can flatten voice. It can simulate expertise. It can make borrowed coherence feel like understanding. A writer can gradually outsource interpretation without fully noticing it.

Those are real concerns. They should be taken seriously.

But that is exactly why the distinctions matter. If we collapse everything into one bucket, we lose the ability to criticize the bad uses accurately and defend the legitimate ones clearly.

AI replacing thought is a problem.

AI helping someone say what they mean is not a problem at all. If anything, it may allow more real thought to become visible. We should be careful not to build a culture that only trusts people who can translate their minds into polished language unaided.

AI helping someone discover what they mean is also not inherently a problem, as long as the writer still owns the judgment, the meaning, and the responsibility for what gets published.

That is the standard I care about.

Not whether a detector found suspicious phrasing.

Perhaps the future of authorship will not be defined by who typed every sentence.

Perhaps it will be defined by who remained responsible for what those sentences ultimately meant.