Can it be understood that a text was prepared by ChatGPT?
ChatGPT can sometimes be used for malicious purposes because the texts it creates are extremely realistic, fluent and logical. Although it is not possible to determine with certainty, there are various methods to determine whether a text was created by artificial intelligence or produced by humans.
ChatGPT is a chatbot that uses artificial intelligence to create texts for users’ requests. The reason why ChatGPT has become so popular lately is that the texts it generates are extremely realistic, fluent and logical. As ChatGPT is experienced by users, we gain new insights into what can be done with this technology.
The fact that the texts generated by ChatGPT cannot be easily distinguished from the human-generated ones causes some ethical problems. For example, in a study whose results were published in bioRxiv, it was determined that scientists who are experts in the field have difficulty distinguishing between the fake research paper abstracts written by ChatGPT and the original abstracts. In addition, the authenticity tests used to examine whether the texts ChatGPT produced contain plagiarism showed that the texts were 100% original content. Some educators say that ChatGPT is used to do homework given by students.
Although the content generated by ChatGPT may seem logical, it may not always be correct. For this reason, it is very important to evaluate the accuracy of ChatGPT’s answers using another reliable source. OpenAI has published a briefing for educators on what to consider when using ChatGPT.
So, is it possible to detect whether a text was created by ChatGPT or by a human?
There are different studies on this subject. For example, in the study, the results of which were published on arXiv, the researchers shared information on the differences between text generated by ChatGPT and text produced by humans. According to this:
While ChatGPT only gives focused answers to the question asked, other topics related to the main theme can be mentioned in the texts produced by humans.
While an objective language based on generally accepted truths is dominant in the texts created by ChatGPT, there can be expressions based on personal judgments in the texts produced by people.
While expressions expressing emotion are less in the answers given by ChatGPT, punctuation marks expressing emotion (for example, exclamation point, ellipsis, etc.) and explanatory sentences in parentheses are used more in texts written by people.
There are also applications that rely on artificial intelligence to detect texts generated by ChatGPT. For example, OpenAI company, which introduced ChatGPT, has released a new classifier application that can distinguish between artificial intelligence-generated texts and human-generated texts. The newly introduced classifier can detect texts generated by artificial intelligence with 26% accuracy. As the length of the text increases, the correct detection rate of the classifier increases. The researchers state that they are continuing to train the classifier and are making the app publicly available to get more feedback. If you want to test the application, you can have the evaluation of a text written to ChatGPT here.