I have thought long and hard about whether I should really add my two cents to the current AI discussion.
As most of you probably know, Substack has integrated the AI detector from Pangram. You can check all stories from July 28th onwards or so.
If the author has not disabled it...
And this is where things actually get funny.
We live in a world that is full of fake and made-up stuff more than ever. So these days it’s quite a bit difficult to find out what is real.
Politicians have been lying all the time. We know that. In some countries, too much lying still has consequences, though. In others, it doesn’t seem to matter anymore.
Good journalism uncovers those lies and points its fingers at them. But this only helps when people actually listen. If they don’t, the liar can go on with his game. And things just get worse, as you can see very clearly these days.
Sometimes journalists get silenced. In autocratic countries, all it takes is a finger snip. In more liberal countries, the game is a little different nowadays. Media is called fake news. Often that’s enough for deflection.
It works because the media got caught here and there spreading false information. Sometimes on purpose, other times by accident or by doing sloppy work. And yes, that should have consequences, too.
The point I want to make here is that we need third-party testing for things that are important to us. Food is an excellent example of that. We want it free of toxins and to be only what’s shown on the ingredients list. Then we can decide whether we want it or not.
Substack offered free third-party testing for articles posted since the end of July. So now we can check whether the text is AI-generated or not. Checking how it’s made, so to speak.
Many people say Pangram is not a reliable detector and that it shows too many false positives. They come up with all sorts of funny stories, such as that Pangram classifies the Declaration of Independence as AI-generated.
Bullshit...
Pangram shows 100% human-written as the result.
Do you need to trust me on that? No, you don’t.
Click the following link to see for yourself:
https://www.pangram.com/history/0daa74a9-831e-49fc-8651-70e2b03d03a9?ucc=UiMOED5oP8Q
Everybody could share the links from the results they got. But they rarely do and present screenshots that don't show the whole story…
The next thing is that people make up fragile test cases. They put a short text block inside. Then they demonstrate how Pangram gives a totally different result after only changing a few words.
While this can be reproduced, it actually proves nothing. Changing a few words in a short text is actually not such a small change percentage-wise. Try the same with longer texts, and you will see that it isn’t that easy to totally flip the classification.
But even if that were the case...
Let’s say I could easily turn a fully AI-generated text into something that Pangram 100% recognises as human-written... then of course I would do those few changes.
The thing is that authors can check their Pangram verdict before publishing a story. So if it just needs a bit of polishing to get it all green, why not do it...
Well, well... people say they wanna stay authentic and not change their writing to please the Pangram algo.
Sounds like a legit point at first. But we writers adapt our writing all the time for those bloody algos.
We know pretty well that we need a good headline. Otherwise, people don’t click it, and the algo does not push the distribution. We also know that we need to avoid certain phrases or words because otherwise Gmail might put our article in the promotions or spam folder where only a few people would open it.
We also want to be found by search engines and accordingly make sure that we please them just a little bit.
So... why in hell should we not want our writing to be classified as 100% human-written...?
Why do some writers switch the detector off and come up with funny excuses?
Should we really trust them when they claim all work was done by themselves?
Would you trust a factory that says the smoke coming out of their chimneys is totally safe? Or would you rather have a look at an independent third-party test?
I am not against using AI. Sometimes we just like fast food, extra creamy chocolate, or whatnot…
I just find it very weird when people deactivate an independent test that you could do with a simple copy and paste anyway. Everyone can sign up for Pangram and have 4 free checks per day, or any other of the better detectors.
But as much as I don’t trust politicians easily, I don’t trust writers who claim they don’t use AI and at the same time deactivate the integrated Pangram test. I also don’t buy into their stories that those detectors don’t work.
I have done my own testing and find them pretty neat.
Long story short, I want to encourage you to use Pangram and other AI detectors to get an idea of what is going on out there.
The more these tests are being used, the higher the entry barriers for all those people who just wanna try their luck. Or, in other words, people being a bit more critical about what they consume could help to reduce the flood wave a bit.
Interesting points. It is a complicated issue, and I too have chosen not to turn AI detection off. If people want to check my writing, that is okay. I write because I like to write, and I do use a grammar checker because I sometimes miss things. I, too, wonder about those writers who turn off the AI checker.