A case where it was suspected that ``I created an essay for a task using AI'' due to false detection of a tool that detects sentences generated by AI, and it became 0 points



ChatGPT, a conversational AI, can generate very natural sentences and can be used in a wide range of fields. On the other hand, New York City has completely banned the use of ChatGPT by students and teachers due to the fear that ChatGPT will infringe on learning and research in terms of education. , and the writing of scientific papers with AI such as ChatGPT is prohibited at international conferences . Under such circumstances, when the teacher used 'a tool to detect sentences generated by AI' for the essay task, the essay written 100% by human power scored 0 as 'written using AI' The student's parent, failnaut, tweeted about the student's experience of becoming a student and has become a hot topic.




ChatGPT's ability to create natural conversational sentences is extremely high, and there are reports of cases such as passing the doctor's license examination and the examination for the master's degree in business administration . In a survey asking about the significance of using ChatGPT in the field of education, about 17% of students at Stanford University in the United States answered that they ``use ChatGPT for assignments or exams.'' Most of the 17% used ChatGPT to generate and consider ideas, but about 5% submitted what ChatGPT output as it was without editing it.

17% of students answered that they use ChatGPT for assignments or exams - GIGAZINE



There is a movement to ban the use of ChatGPT in educational settings, but there are also voices who think that ChatGPT is a useful tool for learning. Attempts are also being made to incorporate education on how to face each other.

An attempt to incorporate ``how to use and how to evaluate'' into student education instead of prohibiting AI-GIGAZINE



Whether it is banning the use of AI tools such as ChatGPT or teaching them how to use them correctly, it is necessary to see through the 'cheating' that is used as is in assignments and papers. However, research results have been reported that the probability that humans can identify sentences written by AI is ``only about 50%, the same level as a coin toss,'' and a tool to detect AI sentences by OpenAI, which developed ChatGPT. Although expectations are high forAI Text Classifier andGPTZero , a tool that can distinguish whether a text is written by a human or an AI, it is still said that it cannot reliably detect content created by AI. increase.

In addition, detection tools may not only fail to detect sentences written by AI, but may also misidentify sentences written by humans as those written by AI. According to what failnaut tweeted on April 11, 2023, when failnaut's son worked on an essay that came out as a school assignment with failnaut's advice, the teacher in charge wrote a sentence generated by AI In the process of checking with the tool whether or not, Mr. failnaut's son's essay seems to have been judged to be 'by AI'. There was no confirmation from the teacher after it was detected as an AI sentence by the tool, and Mr. failnaut's son was given a score of 0 without any explanation. I'm here.

Opinion pointing out the possibility that the submitted essay was judged not to be written by a student because it was 'too well done' as the reason why it was suspected that it was written by AI, and that AI was used. is also commented. In response, failnaut said, ``I said that I supported writing essays, but more specifically, I taught them to write the points they wanted to convey in their own way in the correct structure, and quoted a lot. I only gave advice to avoid using things and supported them to acquire good habits.I didn't do anything to write for them.

Failnaut does not want to directly criticize teachers who use AI detection tools, saying, ``Most of the time, it is used to deceive people, and many children are disillusioned by being falsely accused of cheating many times. I'm really dissatisfied with the fact that such tools continue to be funded and expanded,' he complains about the development of generative AI. In addition, Mr. failnaut also has experience as a lecturer. It speaks to both teachers and concerns.




Failnaut's tweet received a number of comments, stating, 'This seems like one of the important AI ethical considerations. Using AI as a primary tool to deliver life-changing results.' In addition to the opinion that it is not recommended, there are many experiences of encountering similar incidents in university reports and papers and high school assignments. Also, even if a tool is used, it is impossible to completely automate the judgment of whether the text is written by a person or by AI, and in the end, it is necessary to manually check it. It is an ironic incident that the teacher is lazy to let AI do the job of finding students, which is entrusted to AI.'

Mr. failnaut's tweet has become a hot topic on Hacker News, and one user said that in the measures taken at some of the major universities in the UK, ``students use detection tools on their own sentences, I am told that the university encourages me to check so that I will not be suspected. Originally, it was a process to detect and avoid students copying and pasting from reference materials without malicious intent, or unconsciously plagiarizing sentences and expressions. It seems that it will be possible to lower the possibility of being suspected. In addition, it has been pointed out that even with the OpenAI detection tool, only about 26% of AI-created texts can be correctly identified, and there is a 9% chance of erroneously identifying human-created text as AI-created. , There is also an opinion that 'It is strange to use AI tools to identify AI in the first place.'

According to failnaut, the teacher quickly admitted the error and correctly accepted failnaut's son's essay. Failnaut, on the other hand, expresses anger at the possibility that children who are once seen as cheating on their own tasks with AI may lose confidence.




in Software,   Web Service, Posted by log1e_dh