Why it is foolish to think AI can't write like humans

2 points by nitishr 3d ago ↗ HN

Last I counted there were 50+ AI text detectors, not including the humanizers that also sell AI detection (thank you hypocrisy). What's even more surprising is the AI built meters and percentage scores they provide when you enter a text. A load of bollocks if you ask me. Working on AI linguistics if there is something you learn, anything is possible with the right technique and framework. I'm not sure what is more concerning the display of such banality or the actual belief in them, nonetheless, the challenge must be addressed.

There are mostly 2 types of value systems on this, the first is that AI is supreme and will be able to replace human output (or already has) and the second, that AI can never be human. Both are true, to an extent. AI could never be human because what it means to be human is so loosely defined, that there will always be an argument to prove the point for the point's sake. In this context, AI can surely write like humans - and it is mostly humans guiding it in getting there. The point is not whether the writing is human or AI, which can never be certain. It's the grey area of doubt that you can find clues on, for further investigation or reason.

The point of this note is to warn you against looking at symbolic meters and percentage scores. Here's a quick test you can do to evaluate a score (we do this too), copy 200 words from an AI model and test it. Note the percentage, then replace 2 meaningful words at random and note the percentage difference. If 2 words causes a big shift, it's highly sensitive - you can infer that it would not hold up against real world testing. If it doesn't shift much, randomly shuffle the sentences and retry - if it shifts, you have your answer. The reasoning is that if it is truly accurate measure of writing as is, it would give consistent % scores on the same text, no matter how it's sliced. If it's purely for show, it will give wildly different outcomes. Do not base your outcome on running the same text multiple times without changing it, as most detectors will probably use a cached result rather than re-evaluate - lest it makes them look unreliable. As anyone who works with data will tell you, there will always be outliers.

In short, treat detector results as flags and do get carried away by a false sense of UI assurance. Interfaces and widgets look flashy, look underneath. AI can most certainly write like humans (especially when guided by them), it would be foolish to think otherwise.

Taken from https://t9xt.com/writings/why-it-is-foolish-to-think-ai-can-t-write-like-humans

extraextra 3d ago ↗

"if it is truly accurate measure of writing as is, it would give consistent % scores on the same text, no matter how it's sliced"

Sounds to me like a fallacy, even different punctuation can change the meaning of a sentence/paragraph. Let alone reordering sentences randomly.

nitishr 3d ago ↗

I'm curious - why does it sound like a fallacy to you? As far as I am aware, detection models are rarely looking at the underlying meaning, and most claim to scan sentence by sentence and calculate scores on that basis - which is what the note is about. in fact many humanizers rewrite the text so that even humans have difficulty making sense of it, and yet many detectors will give a 100% human result on those humanized texts.

about your note - it's a purely mathematical eq. if some text is 100% AI, how it is shuffled/sliced (get AI to slice/re-order sentences so that it still makes sense; if you want to control for human input) should not change the score significantly.

and - Yes ofcourse, to humans reading - it would make a dramatic difference. But when tools claim 'we give you a sentence level breakdown', it implies a degree of isolated analysis. So without breaking sentence units, it would seem fair to play around to check how much a score changes. For instance, I tried a 75 word piece on a popular detector, the last 25 words came back as AI and first 50 as assisted (roughly). When I clipped the 25 and put them at the front of the paragraph, they came back as human. Not sure if you get what I mean?

Word substitution does change meaning, but here's how I see it - if you change 2/100 words. but the score moves 25%, it sounds like a highly sensitive model, which could also be interpreted as a result based on weak signals. The point here is simply that gimmicks like percentages should be taken as gimmicks, unless you can find more robust ways to substantiate the scores.