news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Since this spring, many college teachers have found thatAIIt has quietly penetrated into the core assessment links of college students' academic performance.

More and more students are using AI to complete their course papers and graduation theses. AI has "unified" the writing style of some people, and big words, clichés, and repetitive words occupy more and more parts of the articles. Many schools have begun to test the AI ​​rate of papers, but absurd scenes are also emerging: those that clearly use AI may not have a high AI rate; those that clearly do not use AI may have a high AI rate.

Detecting the AI ​​rate has become "metaphysics"; reducing the AI ​​rate to a level below that stipulated by schools and teachers is also "metaphysics".

Use AI to write papers, use AI to detect the AI ​​rate of papers, and then use AI to reduce the AI ​​rate. My classmates joked that the world is like a giant robot.

When AI invades graduation thesis, which is supposed to reflect critical thinking and creativity, teachers and students suddenly enter a state of disorder...

Text｜Wang Shuangxing

Editor:Li Tianyu

Invasion

Zhou Ran's graduation thesis was passed with the help of AI.

She is a law student at a second-tier university in Sichuan and will graduate this year. In the spring, after submitting her first draft to her teacher, Zhou Ran received feedback shortly after: "It's very poorly written."

So when revising her manuscript, Zhou Ran tried to use AI. She threw the ideas in her paper to AI, asked it what it thought about them, and then supplemented her first draft. At the same time, she found relevant papers based on AI's answers and then supplemented the ideas in the papers into her own article.

What surprised her most was that AI not only provided theoretical information, but also cases. Even if it did not provide case materials directly, it would provide relevant websites. Zhou Ran used this to add a lot of new content to her paper, making up for the shortcomings of the first draft, which was "too theoretical and lacked practical application."

After the second draft was submitted, the teacher did not say anything except for some minor issues with format and wording. The paper was successfully passed.

Zhou Ran said that during graduation season, almost every classmate around him would use AI to varying degrees to complete their papers, and everyone would also share more useful software with each other in private.

In November 2022, OpenAI launchedChatGPTThis AI tool can communicate by learning and understanding human language, and has strong natural language generation capabilities. In layman's terms, AI can help you write articles.Large ModelFor example, Wen Xin Yi Yan, Tong Yi Qian Wen,KimiAI is getting closer and closer to everyday life, and it is inevitable that it will be used in graduation theses. The 2024 graduates will also become the first group to use AI on a large scale in their theses.

In many university forums, you can always see posts about renting ChatGPT accounts from each other. In other social media, there are also many strategies on how to use AI in graduation thesis, such as "How to write a thesis in 1 hour"GPTInstructions". Some netizens joked, "In the acknowledgments of graduation thesis, AI should be the one to be thanked the most."

Chuchu, a girl from a key university, said that the use of AI has almost become an "open secret" in the school, whether it is a course paper or a graduation thesis.

On weekdays, final exam week is almost the most difficult period of a semester. I have to take several exams in a week, and submit about three or four papers with a total of more than 20,000 words. "For example, for a 6,000-word assignment, it is feasible for me to start from scratch and look up papers one by one, and write bit by bit, but it is not feasible in terms of time. I need a more efficient way of output." Chuchu said.

She first came into contact with AI last year when her classmates gathered together to chat. She lamented that writing a paper was painful. A classmate next to her recommended ChatGPT, saying "GPT doubled my efficiency." She tried it and found it very efficient. She explored and found the domestic AI software that suited her.

Chuchu said that AI can help people write a small article of a few hundred words, but if you want to complete a complete article, it can only help people save time in certain steps. Most of the time, she uses AI as an "advanced browser."

To write a review article, Chuchu often needs to read ten papers in depth and then read twenty or thirty papers. "In the past, it took three or four days to read the materials. It was really a manual labor, and it made me dizzy." Now this work is replaced by AI, which directly selects the most useful materials. At the same time, by uploading a PDF of a paper, it can quickly generate the framework and refined content of the paper, so that you can skip reading without having to read the entire paper. Nowadays, two hours of reading time is enough for the initial stage.

After finishing reading, Chuchu will provide the screened out valid information and her own inspiration to AI, which will generate an article framework. Then she will add flesh and blood to this framework, asking AI what information it can find for more detailed subheadings, and finally complete it with her own ideas. The draft of an article is completed.

"It's not the best, but it's the safest," Chuchu said. However, after using AI more, I found that sometimes the things I wrote by hand also had an "AI flavor" - official, written language style, and habitual "total score" structure.

"Crossing the Tribulation"

College teachers have also gradually felt the invasion of AI. On social media, there are students' usage guides on one side and teachers' observations on the other.

Some teachers questioned, "Many expressions are very strange, I really wonder if it was generated by AI"; others analyzed, "One student wrote it with AI, it was so clumsy that you could tell at a glance, for example, there was not a single paper on the current state of research at home and abroad, it was all empty talk, and the research suggestions were all empty talk that had nothing to do with their own research"; some teachers lamented, "Every year when I supervise graduation thesis, I feel like I'm going through a tribulation, plagiarizing, AI, patchwork, and I think I won't leave any traces"...

Many teachers feel that graduation theses are full of AI flavor and less and less human flavor.

Wang Jingya, a teacher at the School of Humanities of China University of Political Science and Law, found that the biggest feature of AI-generated papers is that they are prone to "empty and clichéd talk". They give a thesis and get three sub-theses. If they ask further questions about one of the sub-theses, they get three more sub-theses. But they often have no research significance.

When Wang Jingya communicates with his colleagues, they also mention that the School of Marxism has become a "hard-hit area" for AI. The same is true for homework and papers in other majors. The more standardized and programmed the content is, the easier it is for students to use AI to complete it.

Like what Professor Wang Jingya encountered, Li Xin, who teaches at a double first-class university, summarized that one can "smell the flavor of AI" from the writing style of the paper - the structure is often very complete, the language is very fluent, and the content is comprehensive, but when it comes to the part where in-depth argumentation is required, it always stays on the surface, repeatedly entangled and smeared.

Li Xin said that when she was marking papers in the past, she would feel angry when she saw incoherent sentences or the use of terms that were close to the students' usual speaking habits but did not conform to academic norms. But now, her first reaction when she sees them is happiness, "at least it means that they wrote it themselves."

The acknowledgments section of a paper is often emotional and "human" in nature, but the reality is that many students have to ask AI to do even this part that requires sincerity and heart, leaving only some official and empty words.

A colleague of Li Xin encountered a similar situation. Seeing that the student's paper acknowledgment was simple and perfunctory, he asked the student to go back and rewrite this part. Unexpectedly, the final paper submitted was still written by AI. Later, the student admitted that he directly asked "how to thank the teacher" and "how to thank the classmates."

The fatal problem with papers controlled by AI is not only that they lack independent thinking and new perspectives, but also that they lack basic respect for academics and facts.

One of Li Xin's students wrote a paper about the dissemination strategy of a certain type of video. In the "Research Status" section, the student wrote: There is a relative lack of research on this topic, and there is no relevant literature. But until the end, Li Xin found that the student still listed a full ten references, which not only contradicted the aforementioned "research status", but the names of these references themselves were too similar: the time was all 2023, the titles all contained common keywords, and each article had two authors, neatly arranged in the format of "XX & XX".

Li Xin noticed something unusual and searched one of the articles. She searched on CNKI but the article didn't exist. She searched on the browser but couldn't find it either. She searched the ten articles one by one and got the same result. Only then did she realize that the ten references were all compiled by AI based on keywords - they were all fake.

Wang Jingya said that the working principle of language logic software is to construct information based on language, but without fact-checking, so there is a risk of "fabrication". In her field of history, AI may also "make up" historical facts, such as "Liu Bannong is a female revolutionary in modern China."

Therefore, when assigning course assignments, she would "give students a heads-up": AI will fabricate some factual content, some of which may not be visible based on the undergraduate students' knowledge reserves, but the teacher can see it at a glance.

Hou Yiling, a teacher at Beijing Foreign Studies University, also posted on Weibo: When I was giving a thesis proposal to graduate students, a colleague in the group said solemnly, "I will offend people today, but I have to say this." (To 6 students) I ran each of your electronic proposals through an AI content detection tool, and it turned out that some of them had AI-generated content as high as 80%! Another colleague and I were shocked. When we looked over, the students' faces changed and they avoided our gaze. It was really worrying. How could students resist that temptation?

But regardless, the wave has come.

"metaphysics"

The country has sensed this trend. On August 28, 2023, the draft of my country's Academic Degree Law was submitted to the Standing Committee of the 14th National People's Congress for deliberation.AIThe corresponding legal responsibilities are stipulated for acts such as ghostwriting of degree theses. It is clearly stated that using AI to ghostwrite degree theses is academic misconduct, and if you are a student, your degree certificate may be revoked.

This spring, a number of universities have also issued guidelines for AI in paper writing. Tianjin University of Science and TechnologyGenerative AIThe school issued a warning to students whose test results exceeded 40% and required them to conduct self-examination and self-correction; Hubei University’s undergraduate school notified that during the review process of undergraduate graduation theses, it would try to add detection of the risks of using generative AI in theses; Fuzhou University said that it would test AI ghostwriting for undergraduate theses of the class of 2024, and use this as a reference for awards and honors…

Zhou Ran's college stipulates that the AI ​​rate cannot exceed 30%, and her thesis advisor has put forward a stricter requirement, which is no more than 20%. Before the test, she was a little nervous because there was a lot of AI content in her thesis. If the number was too high, she would have to spend more time to work on the thesis. But she didn't expect that the number in front of her was only about 5%. She thought it might be related to the fact that she did not completely copy the content provided by AI.

In the eyes of many students, AI rate detection is a "metaphysics". Those who have used it may not have a high rate, while those who have not used it may have a high rate. Zhou Ran's situation belongs to the former, while Chen Lu, a student majoring in journalism and communication, encountered the latter.

The first half of the year just passed was the second semester of Chen Lu's sophomore year. She took a course on Internet and New Media, and the teacher had very strict requirements for the final course paper. Everything was in accordance with the standards of a graduation thesis, with standard format and plagiarism check.

Before submitting the assignment, Chen Lu said that perhaps considering that AI rate detection will be a trend in the future, the teacher suddenly notified that students should submit their papers withAIGCTest report, and the data shall not exceed 15%.

Chen Lu uploaded her paper to the website requested by her teacher. Unexpectedly, a paper written entirely by hand was detected to have an AI rate of over 30%. The paragraphs or sentences marked in red and yellow were typed out by her word for word.

On social media, many people are sharing similar experiences. For example, before graduation, a student received a notice from the school about checking the AI ​​rate. He uploaded his own paper to the website, but ultimately received a result that was higher than the school's requirements - "I typed it word by word and was diagnosed with AI."

No one knows the specific logic of AI detection, and there is no channel for appeal. With graduation approaching, whether one can graduate successfully is at stake. All everyone can do is rewrite word for word the text that was clearly written by themselves but was recognized by AI as being written by AI.

Chen Lu also experienced the same emotional changes, from surprise to speechlessness, and then to helplessness. In the end, she could only face it.

The AI ​​rate is a mystery, and lowering the AI ​​rate is also a mystery.

Chen Lu went online to look up tips and saw many netizens sharing suggestions, such as: exposing some of the "human" shortcomings, such as making the sentences less fluent, or adding more colloquial and down-to-earth content - even though this is contrary to the requirements of the paper itself.

Some people also said that they converted the "judged by AI" paragraph into Estonian, and then converted it into Chinese, and they passed.

Chen Lu didn't try it. She felt a little reluctant to tear the paper she had spent so much time writing into pieces and make it messy.

In fact, there are many such posts on the Internet, but they are mixed with personal suggestions and strategies, as well as various advertisements.Little Red BookOn the platform, someone posted a striking solicitation: "How did I reduce the AI ​​rate from 87% to 2%?" From the introduction, we can see that the AI ​​rate was reduced by using AI tools, "one-click article", but in the comments, the poster mentioned the words "discount coupons" and "tutorials".

Chen Lu has also encountered this. Netizens enthusiastically shared a software that can reduce the AI ​​rate. She followed the steps provided by the other party, but finally found that the software needed to be paid. Chen Lu also tried some free software, but found that it was not easy to use. The logic was broken, the sentences were changed all over the place, and many of them were even incoherent, so she had to give up.

Use AI to write papers, use AI to detect the AI ​​rate of papers, and then use AI to reduce the AI ​​rate. My classmates joked that the world is like a giant robot.

war

During the most demanding period of academic study, lowering the AI ​​rate is tantamount to a war.

It was exam week, and Chen Lu had to prepare for the exam while reducing the AI ​​rate of her course papers. The website provided a free test opportunity every day, and the first thing she did every morning when she woke up was to log in to the website, upload the paper, test it, and then use the free time of the day to fight with the sentences marked in red and yellow.

It wasn't just her. All four of us in the dormitory had AI rates that were all exceeded. We were nailed to our chairs every day, and the air was filled with complaints. When we met classmates on the way to the exam, the way we greeted them became: Is it down?

But what made Chen Lu even more distressed was that her roommate did use AI and also detected an AI rate of more than 60%, but it soon dropped; she did not use it at all, not only was it detected at 30%, but it also took a long time to drop. Not only that, there was a period of time in the middle, it even "risen instead of falling". After her "mental state" was repeatedly hit, she ran to the Internet to post a complaint: Am I an AI? The accompanying picture is three test reports, the data ranged from 28.5% to 26.41%, and then to 27.63%.

After posting the news, she received some suggestions from netizens, as well as some "greetings": Do you need to manually reduce the rate? She found that a new "job" of manually reducing the AI ​​rate has emerged.

Lu Kai, a college student, has been a "foreign aid" to many people involved in this war. Now, he is using his spare time to help others reduce the AI ​​rate to make extra money. April and May are his busiest times. As the defense is approaching, people "place orders" almost every day.

Lu Kai's workflow is that the other party first sends an AI detection report, then prices are set based on the number of words of "suspected AI-generated text" (30 to 40 yuan per thousand words), the other party pays a deposit, Lu Kai rewrites the marked sentences, re-tests the AI ​​rate after completion, takes a screenshot and reports to the other party after meeting the standard, pays the balance, and sends the revised paper. If someone introduces other students, Lu Kai will send a red envelope to express his gratitude.

In the past few months, Lu Kai's orders have ranged from over 100 yuan to around 400 yuan. The most "difficult" order he received was for an article of more than 10,000 words, of which 88% was suspected to be AIGC.

It was clearly an article she wrote, but it was judged to be suspected of AI, and the data had to be lowered. Chen Lu didn't want to pay for such an absurd thing, but she had to face it, and finally she could only continue to "lower" it herself.

Later, she slowly summed up some effective methods, such as: removing the summary before each paragraph, avoiding the stereotyped general-specific structure; changing the sentence structure more often, avoiding repeated words; adding subjects to sentences; not using too many commas, using more periods to break sentences, and dividing into paragraphs; turning short sentences into long sentences, changing "first and second" into "first and second", and changing "and" into "and, and, as well as, etc."...

Although it was just a course paper, Chen Lu took it seriously because of the strict requirements of the teacher. She spent a lot of time on it in the early stage: looking up information, writing, revising and polishing, and she did her best in every link. A lot of time and energy were spent in the process, but the article became visibly better, so she felt it was "worth it". But in the days when the AI ​​rate was reduced, she doubted her life every day, because she knew that what she was doing was meaningless, but she had to do it.

On the ninth morning, I logged into the website, uploaded the paper, and tested it. The data that popped up finally fell within the range specified by the teacher. Chen Lu did not feel the slightest sense of accomplishment or happiness, but only felt relieved, "a hot potato was finally thrown away."

Degeneration

AI is coming with great force, but those who are facing it are still in a state of disorder and exploration.

Some students learned that their schools were checking AI rates, so they worked hard to lower them; some schools did not check, but students were worried that the check would affect their graduation, so they went around to get information every day.

Xinhua Finance once reported that in order to avoid the impact of AI writing, some foreign universities are reducing open-ended assignments to be completed after class and placing more emphasis on classroom assignments, handwritten papers, group assignments and oral exams.

Domestic college teachers also have to face this new challenge. They have to rethink and adjust their training and assessment methods, what is meaningful and what is meaningless, and how to assess the process rather than the results.

In various discussions surrounding AI and graduation theses, some students complained on social platforms: undergraduate students writing papers are nothing more than creating academic garbage; don’t expect a small undergraduate like me to write something profound and academically meaningful; what AI writes is better than what I write.

After the advent of AI, Li Xin also discussed it with his peers.

If a student has ideas about a topic but cannot express them clearly, is it possible to use a software to express them more clearly?

——But after thinking about it, she still felt that being able to clearly express the thinking process and results in a language that everyone can understand is one of the abilities that a person with higher education should possess, and it is also one of the goals that education should cultivate.

If students use AI as an advanced browser, only using it to search for literature, extract key points, and reduce the pressure of reading literature, is this okay?

——Li Xin also thought about it, and the final conclusion is that it is also very important to find and stimulate one's own interests in the process of reading literature, which cannot be replaced by machines. At the same time, screening the information one needs from massive data and finding one's own research clues in various discussions is also a kind of research ability cultivation.

"Traditional techniques are still useful, it's just a matter of efficiency. Once there is a more convenient way, there is no going back. We can only pursue more and more efficiency and convenience, and the losses in this process are difficult to assess," said Li Xin.

Li Xin has been teaching for 15 years and has always been aware of the impact of technology on people.

She recalled that in the past, she occasionally had to maintain order in the classroom because students would chat quietly with each other, and the teacher on the podium would need to add some topics that students were interested in to attract their attention. But now, the class is often dead silent, and almost no one talks, whether in class or after class. Everyone is buried in their phones and tablets. Even telling jokes is difficult to attract students' attention, and getting them to "look up" has become a problem for many teachers.

At the same time, some students who grew up with the Internet and smartphones have difficulty expressing a complete idea in a complete sentence. "They often just blurt out words and phrases." Not to mention a graduation thesis of about 10,000 words, divided into several parts, logically arguing and reasoning, and then explaining a problem clearly. "Some students can no longer do it."

Not long ago, Li Xin read an in-depth report on a dormitory poisoning case. In addition to the incident, it also explored topics such as interpersonal relationships. But she noticed that when the article was circulated on social media, many young people commented: Why is the article so long? Did he do it? "Everyone just wants a straightforward conclusion, but they don't have the ability and patience to understand a relatively complex matter and the subtle and dark parts of human nature in the middle."

Li Xin remembers that some time ago, her colleague, a master's tutor, wrote a lot of comments when correcting a student's paper. Later, the student sent a revised version of the paper, and the colleague found that many places had not been changed. The student was very frank and said, "Teacher, you wrote too much, I can't finish reading it."

"Thinking style and thinking ability need to be trained. If they are not used, they will degenerate. Graduation thesis, in a certain sense, is the training and test of logical ability and thinking ability," Li Xin believes.

The invasion of fragmentation is still ongoing, and AI is coming again. The capabilities of AI are improving bit by bit, but human capabilities cannot deteriorate more and more.

response

There must be a response plan, and it is becoming more and more urgent. New technologies will not stop, and the development of science and technology cannot be stopped. Li Xin said that as a teacher, she is still trying and exploring how to understand and apply it.

This year, the college has a fund for hardware equipment, which is applied for by each major. Li Xin wanted to apply for a Wenxinyiyan account for each teacher so that teachers can understand it first and guide students better. But later, the application was cut off - a computer or a camera is fine, but a virtual account, once it expires, cannot prove where the money was spent during acceptance.

Zhao Siyuan, a professor at Shanghai Jiao Tong University, also talked with his colleagues about the possibility that in the future, some courses will have to return to traditional assessment methods, such as using quizzes and closed-book exams to observe students' learning progress. "It is indeed difficult to know how students completed some assessment content after class."

Zhao Siyuan, who has been engaged in digital humanities research for many years, holds a more positive and open attitude towards the use of AI.

"If a person wants to deal with something, he can find ten thousand ways to do it." Zhao Siyuan said that in the era without AI, people who wanted to perfunctorily write a paper could also find a "gunman". "In every era, there will be people who try to take shortcuts to achieve their goals. This is an eternal social phenomenon. It is just achieved by different technical means in different technological environments." And now in the new technological environment, the guiding role of teachers becomes even more important.

In recent years, Zhao Siyuan has opened a course on paper writing on campus. At the end of 2022, ChatGPT came out, and he realized that it would significantly change the way academic writing works, so starting from 2023, he will take out a class in each semester to discuss with students how to use AI appropriately from the perspective of skills and ethics.

In the past, Zhao Siyuan tried and explored some application scenarios of AI. He felt that AI could replace people to complete some low-difficulty, repetitive tasks, such as making tables and PPTs; it could also help in processing programmed and standardized texts, especially when publishing English papers. Zhao Siyuan once read an article written by a laboratory scholar, who said that the cost of using GPT to polish a paper is about 0.3 cents now, but in the past it might cost hundreds of dollars to do this.

But at the same time, we still need to read the historical materials that need to be read, and do the field research that needs to be done, including running data on the computer and doing experiments at the desk. These are things that people should always do and should do. "Artificial intelligence cannot replace the verification of the authenticity of the real world."

"Every generation has to face the changes brought about by technology," said Zhao Siyuan, but it is only the training method that is changing. Tools will always be an aid to academic training. Whether an article is well written does not depend on what tools are used, but on whether the researcher has thought through his or her research problem.

Wang Jingya, a teacher at the School of Humanities at China University of Political Science and Law, also believes that AI can currently replace some of the work of "hands", but can never replace the "brain".

The language generation mechanism behind AI products such as ChatGPT is a probabilistic model refined from existing corpus. In other words, they can use existing knowledge to answer questions, but it is difficult to create new content and propose new ideas and viewpoints. Wang Jingya said that the most precious thing is often the "jump".

What is the "jump"? She cited Mr. Tian Yuqing's classic work "The Politics of the Eastern Jin Dynasty".

In this book, the author sorted out a large amount of historical materials and constructed a huge network of relationships - the alternation of powerful families, their relationship with imperial power, how they supported a certain emperor, how they conducted family marriages, and how they realized their own family interests.

Wang Jingya said that, in fact, such a network of relationships can be realized today with the help of technology. For example, in the emerging discipline of "digital humanities", there is a very convenient database. "Who is whose in-law, who is whose disciple, who has the same ancestral home, many details are easy for people to overlook, but data can do it and can provide help for research." Wang Jingya said, however, after building the network of relationships, Mr. Tian Yuqing, based on his understanding of culture and sensitivity to politics, achieved that precious "jump": analyzing how the aristocratic families at that time used their own network of relationships to influence the political landscape at that time.

“This is something AI can’t do.”

(Zhou Ran, Chu Chu, Li Xin, Chen Lu, and Lu Kai are pseudonyms in this article)