When she saw the red and yellow graduation thesis in the AI detection system, Shu Ran felt "like crying but not crying." She asked herself in her heart: Do I look so much like AI?
Among Shu Ran's 23,000-word thesis, dozens of paragraphs were detected as "medium risk" and "high risk". Shu Ran explained that this means that although his graduation thesis is original, it cannot meet the school's requirements for passing the review of his thesis due to the "high AI rate".
With the popularity of AI, there are many students who use AI to generate papers when writing them. According to the official website of some universities, for the class of 2024, these schools have issued a notice on AIGC (Generative Artificial Intelligence) testing for the first time, clarifying that students should strictly abide by academic norms and academic ethics when conducting graduation designs (papers). Avoid over-reliance on intelligently generated content and ensure the independence and originality of graduation designs (papers).
The students interviewed all believe that AI testing can indeed prevent students from using AI to write papers to some extent, but the current AI testing function is not yet mature. "It's difficult to understand how it is judged to be written as AI, because even if it is written manually or after modification, it will be detected as AI writing again." Shu Ran said that the result of using technology to sanction technology will lead to the contrary to the requirements of the paper.
For those students who were wrongly judged to write their graduation thesis as AI, the machine gave them a number. They had no way to appeal and could only modify it according to the machine's operating logic. An obvious paradox is that as artificial intelligence becomes more and more like humans, so do the sentences written by humans themselves. At this time, if AI is used to detect whether AI is used, the more likely it is that their paper will be wrongly judged as AI.
Change from "medium risk" to "high risk"
At the end of April this year, Shu Ran, who was about to graduate from college, decided on the title of her graduation thesis after discussion with her mentor.
A month later, Shu Ran began to write the first draft of her thesis. While writing her thesis, she has entered a company as an intern. Every day after work, she lies in front of the computer and racks her brains, writing until midnight before resting. For two days on the weekend, she devoted all her time to writing her thesis.
Because the structure and framework of the paper were discussed with the tutor in advance, Shu Ran's thoughts were clear and smooth during the writing process. In addition, the tutor put forward more detailed requirements for the format of her paper, including the table of contents, fonts, and lines of tables.
About half a month later, she finished writing the first draft of her paper. According to the school's requirements, her graduation thesis must have more than 20,000 words. After revising it over and over again, Shu Ran's paper stayed at 23,000 words.
Another requirement of the school is that in addition to general duplication checking, students need to use an AI tool called "Grid Data" as a paper citation detection system, and the test results must be "low-risk."
Shu Ran learned through online inquiry that this software can provide functions such as paper duplication check, format detection, format correction, and paper management. Paper duplication checking is mainly carried out by comparing it with the journal database to find out the duplication rate of the content in the paper.
Different duplication checking systems have different algorithms and database sizes. For example, the rule for duplication checking on the Internet is 13 characters, that is, 6.5 words will be marked red if they are repeated continuously. The purpose of AI detection is to identify which content is generated by AI and make judgment by analyzing language style, detecting duplicate content, grammar and logic.
A week after writing the first draft of the paper, Shu Ran handed it to her tutor. After the tutor put forward the revision suggestions, she "manually" revised everything from the title to the framework until she was satisfied with herself. Two weeks before her defense, she submitted the paper to an AI tool for testing.
Before Shu Ran submitted it, one of her roommates first conducted an AI test. The result was a duplicate check rate of 13%, which was a "medium risk" for AI detection. Her roommate was not surprised by this result. She told Shu Ran that part of her thesis was written by AI on behalf of her.
At that time, Shu Ran thought it was "quite novel", but she thought to herself that such a paper must have a high repetition rate, and her paper was written word by word. When Shu Ran submitted the paper, the duplicate check rate that appeared in front of her was only 1%, but the AI report was also a "medium risk".
This means that she must reduce her AI rate from "medium risk" to "low risk".
Like Shu Ran, Sichuan college student Lin Tianle's graduation thesis was also judged to be written by AI. Two months before graduation this year, Lin Tianle suddenly received a notice from the school requesting to check the AI rate. The day before the deadline for submitting the paper, he learned that the school required students to use the Vip paper detection system. The system showed that AI language models could be used to detect AIGC to quickly and accurately identify AI-generated content in academic texts, and could effectively identify whether the text was partially or entirely generated by the AI model. The detection results have nothing to do with the quality of the paper, and only indicate the probability that there is a possibility of AI generation in the content fragments in the paper.
Lin Tianle School requires that the AI rate be lower than 30%. If you exceed this ratio, you will not be able to graduate if you fail the blind review of your thesis. After he uploaded the paper, the result was 37%. Lin Tianle said that he wrote all the content labeled AI himself.
Before that, he had already checked it on Zhiwang, and the repetition rate was zero %, so he was more at ease with his paper. Until the day when the first draft was due to be handed in, the teacher suggested that he add new content, but he continued to revise it and did not finalize it until the afternoon. "I feel like I am working so hard and I am still in a hurry before I hand in, but my AI rate is judged to be 37%. I feel quite angry."
Lin Tianle complained about the test results on social platforms. The pictures in this article are all provided by interviewees
Because he had already been recommended for graduate students, Lin Tianle had higher requirements for his thesis. Every morning at 9 o'clock, he would go to the laboratory to check in to write his thesis, writing until eight or nine o'clock in the evening, and sometimes he would stay up late until eleven or twelve o'clock. It lasted for a semester and wrote nearly 40,000 words of paper.
Originally full of confidence, his senior brother, who was studying for a doctoral degree, also praised his paper for its rich content, its beautiful drawings, and the duplicate check rate never exceeded 3%. However, Lin Tianle's mood dropped to rock bottom when he was detected.
Has your AI rate dropped?
At the beginning of revising the paper, Shu Ran thought about it himself according to the prompts to mark red and yellow in the software. For those sentences that are similar to other content,"Look at it a bit like AI", I change them word for word.
She majored in international economics and trade, and her thesis involved conceptual explanations of many professional terms. But what troubled her was that these fixed explanations were most likely to be labeled red by AI systems, and she was unable to "create a new concept."
Her specific modifications are to replace them with synonyms, adjust the order of sentences, or use some more "advanced" words to counter "the mechanical and low-level words given by AI." But she did not expect that the result of the second test would turn into "high risk."
Shu Ran's heart was completely "broken". She doesn't know the algorithms and rules, but only knows that the software will mark different risk segments. The ones marked red are high risks, the ones marked yellow are medium risks, and the ones marked green are low risks. In the first test results, there were 35 paragraphs of "medium risk" in her paper, which became more than 50 paragraphs after being revised.
Shu Ran revised the paper and became "high-risk".
After the paper was judged "high-risk", Shu Ran calmed down for the night. The next day, she began to go online and search everywhere for ways to reduce risks. At first, she tried several AI tools that could reduce AI rates,"but they were of no use."
Later, she found some experience sharing posts about college graduates in reducing AI rates, saying that she would remove the words "first","second", and "harmony" in the thesis, and change sentences and word orders. She revised the paper again in this way, and the paper changed back to "medium risk".
After changing it one day, we are back to the starting point. Shu Ran felt that this modification method was too slow, so she continued to search online. This time, by refining other people's experience in reducing AI rates, she summed up a "rule that she knew very well", that is,"complex becomes simple, honest, proactive becomes passive, simple becomes complex."
There were only two days left before the final version of the paper was submitted, and Shu Ran and her roommates 'papers were judged to be "high risk" or "medium risk." "Some of them use AI to write it, some copy it from other people's papers, and some write it themselves." Shu Ran said that their test results were similar, except that the duplicate check rate was not very high, but the AI rate was very high.
I stayed up until three o'clock in the morning. That night, the most common thing that was said to each other in the dormitory was: "Has your AI rate dropped?"
Although the school provided a free test opportunity, Shu Ran wanted to use it on the final draft because once the test report came out, the system would directly submit it to the teacher. She spent more than 200 yuan on herself and purchased four testing opportunities.
She felt that the paper was like having undergone physical examinations again and again."It's all illnesses that come out." The report is very detailed,"Even if a few words are duplicate, they will be marked red."
During the process of reducing the AI rate, Shu Ran's mind was always the tutor's words: Failure to pass the thesis would affect graduation. There will be random inspections after the defense. If the random inspection fails, the consequences will be serious.
When it was almost dawn, Shu Ran couldn't remember how many times she had edited a complete paper. Upon checking again, although there were still several paragraphs in the report showing "medium risk" and "low risk", the overall result was the three words "zero risk" that she was very much looking forward to.
She breathed a sigh of relief and couldn't wait to share with her classmates her methods to reduce her AI rate. She stayed up late to help her roommate's paper change to "low-risk". "The roommate was very happy and said that you were the god of our dormitory."
But Shu Ran felt that she had also revised her roommate's paper to make it "difficult to read." In order to reduce her AI rate, like other classmates, she no longer cares about the smoothness and smoothness of the paper content. However, in the end, she successfully passed the defense.
Like Shu Ran, Lin Tianle also searched the Internet for various ways to reduce the AI rate. His classmate suggested that he use AI software to reduce his AI rate, but he refused. I rewrote it three times, and finally "exploded" in a fit of anger.
However, the AI rate found in the paper was still 30.13%, and he felt that he might have to postpone his graduation. He remembered that night, almost all the students in the class, like him, stayed up all night to revise their papers because of AI.
At first, the school's notice was to hand in the manuscript before 0:00 that day. Due to the unstable AI testing system, many students failed to hand in it successfully. The school had to postpone the deadline to 12 noon the next day.
Lin Tianle's later experience in reducing the AI rate was to use less conjunctives, use more commas in sentences, and multiple short sentences share the same subject.
In order to be faster and more efficient, Lin Tianle deleted all the parts that were tested and marked with red and directly rewritten them again. After correcting it three times, there were still red marks in the paper. He continued to rewrite it again, but in the end, the AI rate only dropped by less than 7%.
The deadline for the manuscript became increasingly tight, and he could only continue to delete the thousands of words marked in red. He knew that this had a certain impact on the content of the thesis."The impression is not as good as before, but there is no way to at least graduate smoothly."
"Deliberately exposing human shortcomings"
When a classmate asked Shu Ran how to reduce her AI rate, she half-jokingly said that it was actually "deliberately exposing some small human shortcomings", such as writing less smoothly.
The joy of successfully reducing the AI rate did not last for a few minutes, and Shu Ran began to feel lost. The paper that she wrote with all her efforts was "mended" and finally became "fragmented" and did not achieve the goal she wanted.
This also made her think that when AI becomes more and more intelligent, what it writes is "very advanced, very professional, and getting closer to humans, and then it uses an AI detection system to let us reduce the AI rate. Isn't that making ourselves less human?" Shu Ran feels that this is a paradox.
Shu Ran's paper was marked yellow by AI before it was revised.
After Shu Ran's paper was revised, it was no longer marked yellow, but the sentences became illogical.
Regarding the behavior of being wrongly judged, the common question of the college students interviewed is, how is the procedure determined? "Many people use AI to make their spoken language academic. Now in order to avoid censorship, they deliberately change their academic language to be more colloquial." said Gu Siqi, a senior graduate of a university in Zhejiang.
In December last year, Gu Siqi completed the opening report of her graduation thesis. In February this year, she began writing a first draft. It took about two weeks to review the materials, conceive and write the manuscript, and she handed the first draft of the paper to the teacher. After that, it will be revised according to the teacher's suggestions, and an undergraduate graduation thesis will be completed.
The school where Gu Siqi attended also has new regulations this year. Different from the paper version of previous years, students need to upload their papers through the HowNet system. After the finalization was finalized, it was in May that the school notified students that they would check the AI rate. Gu Siqi thought to herself, I have written it all, so I won't change it again. "The teacher only said that if the evaluated papers (AI rate) need to be below 10%, but did not say how much the AI rate of the qualified papers should be lower." Gu Siqi recalled.
After the paper was uploaded to the testing system, Gu Siqi discovered that a certain paragraph she wrote,"a very logical conclusion," was identified as written by AI. Although there is no specific ratio requirement, the teacher said to try to reduce the AI rate of the paper. Before dropping, the AI rate of Gu Siqi's paper was 38%.
So she had to remove the logical words to make the correlation between the sentences weaker, either change the order of paragraphs, or compress several paragraphs into one paragraph. After the disruption,"it looks more like it was written by a machine." However, after half a day of modification, the AI rate of the paper did drop.
However, Gu Siqi felt that the "highlights" in her paper eventually became "plain". "This made me feel very uncomfortable when reading the paper."
After she heard that AI could be used to reduce AI rates, she also tried using ChatGPT and other software. At that time, she wanted to clear out the related words, but the final result was "completely irrelevant" to what she wanted to write.
Gu Siqi felt,"There is a certain gap between the logic of AI and the logic of human beings. It will carry some people's thoughts, but it's not as smooth as us people write and it looks awkward." She said that AI will only grab some keywords and identify whether the paper was written by AI. Therefore, using AI to change the AI rate seems contradictory.
After repeated revisions, Lin Tianle has lowered his "academic requirements." Forcing changes in the logic of writing also made him feel "confused". "Connectives that have been used since childhood can make articles smoother, but they can't be used now. They seem to have become exclusive to AI."
Although Lin Tianle's thesis was rated as an excellent graduation thesis as he wished, he still had a feeling of injustice that he had nowhere to complain about. "I haven't written by AI at all. Do I want people to admit something they haven't done?" He felt sad for several days.
Technology "sanctions" technology
Before and after the production and submission of a paper, AI is used as an auxiliary tool, then used as a system to detect AI ghosting, and then became a tool to reduce AI rates."AI dominates everything," Lin Tianle said.
Lin Tianle's major is automation. Sometimes he will use AI to ask about the relevant meaning of a certain code."AI will give more patient and detailed answers than humans." He recognizes AI as a practical tool. However, regarding the AI rate of checking papers through AI, he felt that the testing standards and accuracy were not convincing.
Lin Tianle once told his thesis instructor his confusion. The teacher felt that the detection system was indeed not very smart at present, but the school had relevant regulations. There was no other way than to encourage him to speed up his revisions.
This year is the first time that Lin Tianle studied has conducted an AI testing graduation thesis. Although the system is still in the public beta stage, it has already begun to be used. "It wasn't said that the Degrees Law would test AI, but in the end the law deleted the AI clause. Because it can only be judged based on simple wording, and it is suspected."
He refers to the "Academic Degree Law of the People's Republic of China (Draft)" reviewed in August 2023, which clarified that academic misconduct such as plagiarism, plagiarism, forgery, data fraud, and artificial intelligence ghosting shall be revoked by the degree-awarding unit after deliberation and decision of the degree evaluation committee.
However, in the Academic Degree Law that was finally promulgated in April this year, artificial intelligence ghosting was not included in the consideration of academic misconduct. "However, some schools have already started to conduct AI testing." A teacher from a university in Chongqing said that the introduction of the "AIGC (Generative Artificial Intelligence) Testing Service System" in colleges and universities is mostly due to academic integrity considerations to ensure the originality of papers., and to resist the behavior of artificial intelligence to substitute papers.
Notices from some schools show that artificial intelligence tools can only be used for auxiliary work such as document retrieval and data processing, and are strictly prohibited from being directly used in paper writing. However, different schools have different requirements for the proportion of content generated by artificial intelligence.
In the process of desperately reducing the AI rate, Gu Siqi had the same confusion: "Is AI's judgment accurate and reliable?"
She heard a classmate say that they thanked a certain teacher in their thanks, which was also marked as AI writing. She didn't know why her paper was sentenced to be written by AI, and she had no way to change it. After the initial change was completed, it only dropped 1%.
There were a few paragraphs in the end that she really didn't know how to modify and couldn't delete. So she used the method of weight reduction, first translating into English and then back to Chinese, and the AI rate finally dropped to 15% of the school's required. In order to reduce her AI rate, some classmates around her first translated Chinese into Estonia and then into Chinese.
There was a classmate in Gu Siqi's group whose introduction and thanks were written by AI. After sending them directly to the teacher, the teacher "criticized" the students,"Such behavior must be academic misconduct. But if your own ideas are greater than the participation of artificial intelligence, AI is just a tool."
After the school asked to start checking the AI rate, Gu Siqi communicated with her mentor. Her tutor usually likes to use AI to assist teaching and does not restrict students from using AI. The tutor feels that using AI can help students expand their thinking."For example, AI gives three ideas. If you can expand four or five ideas beyond AI's ideas, you will have the initiative to interact with AI."
In the future, Gu Siqi plans to go to graduate school. She believes that future papers may be tested by AI, and testing is necessary,"because there are indeed many people who use AI directly to write." However, the content she wrote herself was also judged to be written by AI, indicating that the system is not yet complete.
In June, Shu Ran passed the thesis defense and successfully obtained her diploma. Her graduation thesis was lying on her computer. Occasionally, she would recall the "painful process" of reducing the AI rate of her thesis, as if she had just completed a task "muddleheaded".
She still knew nothing about the system that judged her to be ghosted.
(In order to protect the privacy of interviewees, all characters in the article are pseudonyms)