Achievement Unlocked: 100% AI

In an absurd turn of events, one of my most popular posts ever here was one that was completely written by AI. Apparently people are entirely capable of engaging with ideas despite the presence of AI-assistance!

So I’m going to continue to post here, but I’m also starting a new game.

Most (though not all) articles on my Substack are written with AI as a writing partner. I am not transparent about exactly how I use it. My ideas come first and AI helps me express them more clearly, reorganise them, and improve the language.

So, from now on, instead of adjusting my writing to lower that Pangram score, I’ll try and do the opposite.

This is the challenge:

Can I get Pangram to label my articles as 100% AI-written while every argument, opinion and conclusion is still entirely mine?

If I manage it, I consider that an achievement unlocked.

Not because I’ve proved Pangram works.

Quite the opposite.

I’ve shown that a detector score tells you remarkably little about who did the thinking.

This is the same argument I make in my work on assessment and authorship.

A valid assessment depends on the quality of the evidence supporting the claim you’re trying to make. Pangram analyses patterns in language. From those patterns, people often jump to a much larger conclusion: who authored the work? who did the thinking?

That’s a validity problem.

A detector can provide evidence that AI may have influenced the wording.

It cannot, from the text alone, provide valid evidence about where the ideas came from, who constructed the argument, who evaluated the evidence, or how they changed their mind along the way.

So I’ve decided not to play the detector-avoidance game.

I’m not interested in making my writing look less AI-assisted than it is.

I’d much rather just invite you to judge what actually matters to you.

Read the ideas.

Disagree with the arguments.

Question the evidence.

But don’t confuse a statistical classifier with evidence of authorship.

Pangram: 🎯 (score varies by article)

Ideas: Kelly

Reasoning: Kelly

Judgement: Kelly

Expression & language refinement: AI

Validity of inference that Pangram can make about authorship: ❌ Not demonstrated.

Pangram detected patterns in writing.

It did not determine who did the thinking.