Welcome the new Substack with AI Detection:

there is a trap with the new AI detection on Substack a writer will begin in his own voice

the detector objects

so he flattens the prose

roughens the syntax

and makes the work look less like himself

until the machine approves it

at that point

the detector is no longer evaluating the writing

it is directing it

the medium is the message

now the medium has become the editior

the score has become the standard

and the writer has surrendered his orientation to the instrument meant to measure it

we built an AI system to catch people using AI

then we let it teach is how to sound human

"Marshall McLuhan, please call your office!"

There is a trap in AI detection on Substack.

A writer begins with his own voice. The detector objects. So he flattens the prose, roughens the syntax, and makes the work less like himself until the machine approves it.

At that point, the detector is no longer evaluating the writing.

It is directing it.

The medium is the message.

Now the medium has become the editor. The score has become the standard. And the writer has surrendered his orientation to the instrument meant to measure it.

We built an AI system to catch people using AI.

Then we let it teach us how to sound human.

Marshall McLuhan, please call your office.

**The three screenshots:

1. What I wrote with no punctuation: Score: Fully Human-Written

2. What I wrote with caps and punctuation: Score: Fully AI-Assisted Text

3. Both of them combined in succession: Score: Fully Human-Written