In my last article I wrote about the IA chatbox and the consequences that it can bring to STEEM. Especially because I had readed a new that said how many users were using chatgpt for write books and sell it in Amazon Kindle.
I was concerned that we would not be able to differentiate between a text written by a human and one written by an artificial intelligence. And I put my ideas into practice. I think that if we accept the use of AI as a tool and users declare which of their texts were generated by these we could create a healthy environment in conjunction with new technologies (although AI-generated text should not be treated as the author's original and therefore should not represent the majority of the article).
The real problem occurs when a user publishes an article whose content has been generated entirely or mostly by Artificial Intelligence, but the user does not declare it but presents the product as coming from his intellectual capacity trying to deceive the reader, desecrating the essence of STEEM and being rewarded without having made the slightest effort.
While these thoughts were still fresh in my head I happened to get a mention from a user promoting his new book which he is selling on Amazon Kindle.... If I just read that the new trend is to use AI to write books and sell them on Kindle, how could I not be suspicious of this user's sincerity? even if he might be totally innocent, at least I should investigate!
That's what I've done, I'll tell you how and the tools I found to do it, how to use them correctly and the probability that an innocent person is guilty!
Pixabay Testing the available tools |
|---|
At first I thought it was wonderful that the user was using his blog to promote his products (this is one of the uses for STEEM that we have exploited very little) and I think it would have been more perfect if he had offered his products with availability to be paid in STEEM. But before leaving a comment I had to do my research not only for this particular author but to find a way to protect us from all those who want to try to cheat us now and in the future.
And with the help of Google I found an article that presented 3 tools created to detect AI-generated texts with certain probability. These are Writer, GPTZero and ChatGPT Detector .
There were the tools, now I just had to test their effectiveness and the best way to do it was with content whose origin was clear to me, my own content... So I used my previous article in which at the end I used an AI-generated paragraph (declared as such) to compare the results of the tools when I used paragraphs generated by me and when I used this AI-generated paragraph, that way I could see the reliability of the results.
All 3 tools detected the paragraphs written by me as "human generated content". So far so good and it was time to test with the AI-generated content you.com to which both Writer and GPTZero detected the text as human-generated.
But the third tool did detect that the text was generated by an AI so this tool passed the test with 100% effectiveness.
Interrogating the suspect |
|---|
It seems that I have one functional tool and two that may or may not work, at least sometime they could detect something so I didn't rule them out at all. If that user was using AI to write his books the probability that he would do the same to write his steemit article was very high. For the verification to occur without errors you must examine one paragraph at a time and that's what I did.
More than 50% of the paragraphs in the article were AI-generated.... 😥
Don't be quick to point fingers at someone you suspect. |
|---|
For a moment I thought that maybe there was a probability that the user has used an AI translator to write his article in a language he does not master and I have learned from the DU community that if the writer is using a translator the final article is written with the translator's words and although the message may keep its essence the expression is not the one it would have if the author wrote in his native language.
So maybe an AI translator chose the words under some algorithm and that is why the tools detected that the paragraphs came from an AI.
To check it I decided to write this article using a translator Deepl, that way I could check the translated paragraphs with the tools I had just found.
Indeed the first two detected some paragraphs as AI-generated but they were the same ones that made the mistake of detecting the AI-generated paragraph as human-generated so that makes me discard them as definitive verification.
However the ChatGPT Detector tool showed that every single paragraph I write here using the translator has been generated by a human, which it did not do with the paragraphs of our friend the amazon seller.
This situation is uncomfortable for me but it has mentioned people that I usually read and identify as "good people" and even with all the above I do not assume that the user is not genuine, however I invite these friends not to trust so soon, also I leave the tool ChatGPT Detector so that it can be used by content verifiers, in this way they could check the possible introduction of non-genuine text in the articles.
Kindle is filling up with AI written books, there are even youtube tutorials on how to do it. Let's take care of our STEEM Blockchain, if we will welcome AI's as a tool let's do this in a conscious way always respecting those who work hard to generate their content 100% genuine, not in a malicious or misleading way.
I think this article could be of particular interest to: @patjewell @damithudaya @weisser-rabe @dove11 @waqarahmadshah
Thank you for reading my article, I hope you enjoyed it. Stay tuned for comments!
