30 July, 2025

Danger, danger, danger ... AI Detected

 

AI Detected ...

My thoughts for today – relates to AI being used by students to write their term papers etc. I have been reading from many (particularly those from academia cry foul with regards to how AI is destroying the minds of our youth. Their claim – many today do not have the ability to do critical thinking as they have become dependent on AI to do the work for them. Final term/year papers are written by AIs and the students have no clear idea of the content of their “own” writing.

While I am sure there are those who are doing that – but to argue that AI is destroying our youth because they become overly dependent on AI to do their work is, I think, too huge a jump for me.

Wanting to look deeper into this, I decided to go on Google (yes, I am still a little “old” fashioned) and enquired about applications or software that claim to be able to detect if a document is the work of AI or if it was developed by a human. I found Sidek******* as one of the “better” tools for this purpose.

As my latest book was just released, I thought I would use that to test it on this particular program. And the results it returned?

WOW! It came back with: 80% AI generated and 20% by human. How can that be when I wrote the book myself. I will confess – I did use AI to assist me. The part that AI assisted was primarily to check on my grammar, coherence of content, consistency, flow, etc. So, the final text was greatly polished I am sure, and in many ways probably presented more nicely than I would/could have. However, I did read through any suggestions AI made several times to make sure the tone and intent remained unchanged. There were times I did not like how AI presented, and I kept my original or brushed that sentence or paragraph up myself. But again, full disclosure, there were more times than not – I did prefer AI’s words and structure. But 80% credit goes to AI, and 20% considered my work? Are you kidding me?

So I decided to take a book which was written by an international (famous) author. This is one of my favorite books growing up and was published in 1998 - at least the edition I have was published in 1998 - but no telling when the first edition was published. I am sure it was way before 1998 because I read it probably in 1984 or 1985. And that book wasn’t even brand new then.  And since there were no AI aided authors then, I thought this would be safe "test." Results: 82% AI, 18% human. WOW!!! I guess my book is not so bad then.

I returned to my AI (yes, this time I got a little more advance and left Google for Perplexity) and expressed my surprise with what I had encountered. AI’s response:

Your experience with Sidek***** and other AI detection tools reflects a common and well-recognized limitation of current AI detection technology: they often produce false positives and cannot reliably distinguish polished human writing from AI-assisted writing or even purely human writing, especially when the text is highly edited and coherent.

Here are the key reasons why this happens, based on recent assessments of AI detectors in 2025:

  • Detection is probabilistic, not definitive: Tools like Sidekicker analyze patterns in writing style, coherence, repetition, and linguistic features typical of AI. But if your book was written by you and only lightly polished by AI for grammar and flow, the text may still carry features (like high coherence and typical phrasing) that the detector associates with AI. Hence, it may flag a high AI percentage despite your genuine authorship.
  • False positives are common: Studies show some AI detectors mark up to 30% false positives—flagging perfectly human-written, even famous published works as AI-generated. Your test with a book from 1998 also returning 82% AI-generated is a strong example of this issue.
  • AI detectors struggle with polished or "humanized" AI text: Your editing and tone-checking probably "humanized" the AI grammar fixes, but left the overall style sleek and consistent—which can mimic AI writing patterns.
  • No current AI detector guarantees 100% accuracy. Leading tools like Originality.AI (85% accuracy), GPTZero, Quillbot, and Sidekicker achieve high accuracy on average but still cannot definitively label every individual text correctly. They provide a confidence score or probability rather than absolute proof.
  • Famous published texts and professionally edited human writing often test as AI due to their polished, consistent language patterns, free from errors or creative inconsistencies typical of casual writing.

A recent comparison of top tools shows that different detectors vary in sensitivity and tolerance; some are more prone to flagging human writing as AI. Sidek*****  advertises itself as reliable and used professionally, but no tool is infallible.

My take away – can’t trust these AI detector softwares. If we are guilty for using an AI to check on our “presentation” skills and how we format (the layout) of our book or our work – then wouldn’t published authors from the past be guilty of not being authentic when they engaged professional editors to proofread, review, and polish their work before they are published? I, for one, am very glad I have a tool that can help me improve the readability of my work at no monetary cost (as compared to engaging a human editor to do that). Who knows, if I went to a human editor, he/she might be using an AI to review and tweak my work anyway. Don’t be misled into thinking with an AI it’s all a breeze to do it either. There were many occasions when “we” had to go back and forth “arguing” about what I want to say and what was being re-represented in the editing process. But, all said and done – and at the end of the day – I am thankful for the technology. (We won’t go into the moral debate here on whether AI is taking away the jobs of humans – in this case perhaps the editor. That’s a topic for another day.

 


Final note:
I just ran the entire writing above (which was 100% me with no AI assistance) minus response from AI I attached above. Expecting this time it would give me at least an 80-90% human since I didn’t even have AI proofread my writing. Want to know the results?
Here it is:
88% AI written, and 12% human-written. LMOA!!!!

No comments: