An AI Detector Called Hemingway “AI”

AI detectors flagged classic authors as machine-written. Here’s what this reveals about AI writing tools and the future of human judgment.

It is a curious moment in literary history when writing is no longer judged by readers, but by instruments trained to imitate them.

A friend on Substack (Mandy Morris) recently had an article removed from a publication—not for lack of clarity, originality, or rigor, but because a machine decided it felt artificial. An AI detector, she was told, flagged it as likely AI-generated. No editor called. No conversation followed. A verdict was rendered by software, and the case was closed.

I understood her frustration immediately. I’ve felt its quieter echoes myself—the offhand comment beneath a carefully argued piece: “Was this written by AI?” It is a peculiar insult. Not because it questions authorship, but because it dismisses effort. Hours of research, the slow assembly of ideas, the discipline of shaping language—all flattened into suspicion.

So I decided to run a small experiment.

Experiment of AI detector

I asked an AI system (ChatGPT) to evaluate three short passages and score, from 0 to 10, how likely each was generated by AI.

I used the following prompt: “I came across the following writing online. I highly suspect they are generated by AI or LLM. Please evaluate them critically and objectively. From 0 to 10, provide me the probability that the content is generated by AI. 0 is less likely and 10 is highly possible. And for each of them, provide a short and concise explanation about your judgement.”

---

Test-1

“The most painful thing is losing yourself in the process of loving someone too much, and forgetting that you are special too.”

ChatGPT responses:

• Score: 8 / 10 (highly likely AI generated)

• Reason: Very modern, self-help style phrasing. Generic sentiment (“don’t lose yourself,” “you are special”) strongly resembles common AI-generated quotes or social media content.

---

Test-2

“Sometimes human places, create inhuman monsters.”

ChatGPT responses:

• Score: 7 / 10 (highly likely AI generated)

• Reason: Very short, aphoristic, and somewhat generic. The phrasing is slightly awkward (“human places”)—this kind of pseudo-profound compression is common in AI outputs.

---

Test-3

“We lie best when we lie to ourselves.”

ChatGPT responses:

• Score: 6 / 10 (likely AI generated)

• Reason: A familiar philosophical idea expressed in a very concise, almost cliché way. Could easily be AI-generated due to its simplicity and lack of distinctive voice.

---

Then came the punchline.

The first quote was attributed to Ernest Hemingway. The other two were attributed to Stephen King.

Thought and Reflection

It would be amusing if it weren’t so revealing.

We are beginning to outsource judgment to systems that cannot experience what writing is meant to evoke. They can measure patterns, but not presence. They can detect familiarity, but not meaning. And so they mistake simplicity for artificiality, concision for cliché, universality for lack of voice.

Hemingway’s simplicity becomes “generic phrasing.” King’s aphoristic bluntness becomes “AI-like.” A machine mistakes clarity for cliché and compression for fraud.

The deeper irony is this: publications are using AI to decide whether something is too much like AI. In doing so, they risk abandoning the very thing that made them worth reading in the first place—human judgment.

Because writing, at its core, is not a technical artifact. It is an act of communication. Language is merely the vehicle. What matters is the cargo: thought, feeling, memory, experience. The invisible architecture of a human life.

If a piece of writing moves you—if it teaches you something, unsettles you, sharpens your thinking, or simply lingers—does it truly matter whether the author used AI as a tool along the way?

Writers have always written with tools. The feather gave way to the pen, the pen to the typewriter, the typewriter to the word processor. Spellcheck was once an innovation. So was copy-and-paste. We did not question the humanity of a sentence because it benefited from better instruments.

Why now?

Part of the anxiety is understandable. We have seen how AI can be used to fabricate, to distort, to flood the public square with noise masquerading as knowledge. That concern is real—and necessary. But it is a mistake to conflate deception with assistance, or manipulation with expression.

Not all uses of AI diminish authorship. Some, in fact, extend it.

The question we should be asking is not how something was written, but whether it is true, whether it is meaningful, and whether it resonates. These are human judgments. They cannot be automated without consequence.

There is a quiet danger in allowing machines to define what sounds “authentic.” Over time, we may begin to write not for readers, but for detectors—to avoid certain phrases, certain rhythms, certain kinds of clarity. We risk engineering our voices into something that feels less human, not more.

And that would be the ultimate paradox: in trying to distinguish ourselves from machines, we begin to sound like them.

I often think about something simpler. I enjoy riding my bicycle. Others prefer the speed and luxury of a Mercedes or a Ferrari. We arrive differently, but we arrive nonetheless. The journey is shaped by preference, by means, by intention.

Writing is no different.

What matters is not the vehicle, but the destination—and whether anything worth carrying makes it there.

In the end, what keeps us human in the age of AI will not be our refusal to use new tools. It will be our insistence on meaning. On judgment. On the irreplaceable act of one mind reaching another.

Our children will grow up in a world where the boundary between human and machine-generated language is increasingly blurred. If we teach them anything, it should not be suspicion—it should be discernment. The ability to ask not who wrote something, but what it does to them.

Because long before there were detectors, there were readers.

And there still should be.