AI examiners force students to cheat

Bian Ce Xiaoxiao from Aofei Temple
Qubit Report | Public Account QbitAI

When you were in school, your teachers must have taught you: If you encounter a question you don't know the answer to, you should write a few sentences, even if it's just random. Maybe the examiner will be in a good mood and give you some points.

I don’t know whether the teacher will give you a grade, but AI will give you a grade, maybe even a full grade, even if you know nothing.

Recently, an AI teaching platform that claimed to serve 20,000 schools in the United States crashed. It was discovered that students could easily pass the exam by taking advantage of its loopholes. This was simply a "good news for poor students."

Rollover process

The cause of the incident is as follows:

One day, Dana, an American parent, noticed her seventh-grade son suddenly burst into tears because he had only scored 50% on a history test.

This means that he must answer all the remaining questions correctly, otherwise he will fail.

The question asks: How did Constantinople's geographical location help the Byzantine Empire become rich and powerful?

The child's answer was basically correct. (Please ignore the last sentence; it wasn't written by the child, for reasons that will become clear later.)

At first, Dana wondered if the teacher who graded the papers was too harsh. She consoled her son like this.

But the child said that he got the score one second after submitting the answer. So Dana realized that the test paper should be graded by AI instead of humans .

Dana has some understanding of technology and believes that the AI ​​system is likely grading based on keywords. To verify her idea, she did her son's homework again.

She tried to include "wealth, caravans, China, India" in her answer, which is the last sentence in the picture above. In short, she just threw in every keyword she could think of.

Miraculously, she scored full marks on this question , even though there was no connection between the key words.

Word Salad - Dana has found a great way to get high scores by combining all possible keywords together, so she'll always get the right one. This method is especially effective for short essay questions.

Of course, this is just Dana's guess. In fact, the company has already admitted in its official help center: We score based on the percentage of keywords .

If there is no keyword, you will get 0 points. If there is one keyword, you will get a little point. If all keywords are hit, you will get 100 points.

Cheating tricks to bypass AI

The company that provides this AI scoring technology is called Edgenuity .

The COVID-19 pandemic has prevented many schools in the United States from opening, and students can only study at home, so many schools have used Edgenuity's remote teaching and testing solutions.

Edgenuity's AI scoring problems have existed for a long time, but were only recently discovered by foreign media.

A week ago, some netizens complained that they only got 0 points, and many people pointed out the fact that Edgenuity relies on keyword scoring.

A high school student who used the system many years ago said he discovered this cheating method back then. When he had no idea how to answer a question, he would fill in the answer box with various words, and it "often worked."

Other students discovered that they could earn points by pasting questions into their answers, as questions often contain key words.

I remember my teacher once saying, if you encounter a question you don't know how to answer, just copy it down . And it's true! AI is actually easier to talk to than human examiners.

Some students told reporters that they used this technique last semester and got good scores every time.

If you search for Edgenuity on Google, you will find that "how to cheat in Edgenuity" has always been a hot search.

On the Edgenuity forum on Reddit, students openly discussed how to cheat. Edgenuity, which claims to be an anti-cheating and automatic grading system, is actually riddled with vulnerabilities.

After clicking this link, I was really opened up to a new world. It turns out that American middle school students have done in-depth research on how to cheat in AI systems.

Doesn't Edgenuity prevent students from having to browse the web for answers? No problem.

Set the browser to delete cache and cookies when closing, open another browser window to search for answers, and it's so easy.

Actually, it doesn’t have to be so troublesome. Nowadays, smartphones are so advanced that you can just go to Brainly (similar to the homework helper in China) to find the answer. After all, the phone and computer are separate.

In other words, Edgenuity must be combined with human teachers, who need humans to help supervise exams online and review the results of the exams in order to be effective.

Apparently, some teachers in the United States are being lazy.

More than 20,000 American schools use

So, what is the origin of Edgenuity, the company that produces "artificial intelligence"?

In fact, Edgenuity, founded in 1998, is a leading provider of online courses and teaching services to the K-12 market in the United States.

After Edgenuity was acquired by Weld North for $50 million in 2011, more than 20,000 schools now use this teaching platform, including 20 of the top 25 school districts in the United States.

In other words, Edgenuity can be found in 80% of the regions with the best educational resources in the United States.

However, the original Edgenuity was primarily an educational content provider, but it acquired an educational software provider called Compass Learning in 2016.

Compass Learning was founded in 1964 and is mainly responsible for developing learning assessment software for elementary and middle schools. When it was acquired, it was already being used by more than 2 million American students and more than 10,000 American schools.

Some time ago, a netizen recalled the fear of being dominated by Compass Learning-related software, which received a strong response from netizens on social media.

Now due to the epidemic, most American schools have turned their attention to online education courses, and some schools have even directly outsourced teaching and testing to online education companies such as Edgenuity.

In response to this matter, Edgenuity said that AI grading will only provide "algorithmic guidance" and cannot determine students' course grades.

Edgenuity does not use algorithms to replace teacher grading; it only provides a recommended grade. Teachers have the ability to accept or reject the recommended grade. AI does not grade tests or determine student test scores.

But even so, Edgenuity's influence is still expanding in the United States.

According to PR Newswire, Edgenuity recently won the 2020 Tech & Learning Awards of Excellence , the EdTech Cool Tool Awards, one of the most representative EdTech awards in global education , the CODiE Awards , the only core award of the Software and Information Industry Association (SIIA) in the United States , and the Tech Edvocate Awards .

These “expert-assessed” technical expertise awards make Edgenuity an even more influential company in online and K-12 education.

AI grading teachers are also available in China

In fact, AI graders are not only found abroad.

As early as 2017, Alibaba launched an AI system for marking Chinese test papers, producing "the world's first essay graded by a robot."

According to Sohu, at the International College of Zhejiang International Studies University, Alibaba AI graded Chinese test papers for 11 international students from Russia, South Korea and other countries. "It usually only takes 40 seconds to grade a test paper."

He reads even faster than the teachers who mark college entrance examination essays.

However, it seems that the main criterion for AI to revise essays is focused on "correcting grammar."

For example, in an essay titled "Please write about your hobbies," one student mentioned Chinese as their hobby and wrote, "The difficulty of Chinese isn't the characters, but the fickleness of it, like a girlfriend. I won't give up until the Chinese give up, too."

△ Picture from Sohu

Regarding these two sentences, a Chinese professor said: "I've taught Chinese for 14 years, and if I didn't look closely, I wouldn't have noticed the problem. But the facts show that both sentences are grammatically incorrect, with errors of mixing and repetition."

It is reported that the main purpose of Alibaba developing this marking system is to teach foreigners to learn Chinese, so there will be strict requirements in grammar.

If Alibaba's marking system is still a small-scale experiment, iFLYTEK has already applied it to actual exams.

It is understood that as early as 2016, my country's Education Examination Center and iFLYTEK established a joint laboratory to jointly carry out research on artificial intelligence technology in paper marking, question setting, examination evaluation and analysis.

In 2017, Xiangyang, Hubei Province officially introduced this machine scoring system as an auxiliary scoring method in the marking of the high school entrance examination.

It is understood that this system is designed and developed based on iFlytek's core artificial intelligence technologies such as handwriting recognition, natural language understanding, and intelligent evaluation, which have independent intellectual property rights. It can detect blank questions in all question types except multiple-choice questions.

Not only that, it can also correct subjective questions and conduct computer intelligent scoring for Chinese composition, English composition, English translation, short-answer questions in liberal arts, and English fill-in-the-blank questions.

At the same time, it can also effectively detect abnormal answers in Chinese and English compositions where candidates plagiarize model essays, effectively reducing the workload of manual grading of subjective questions.

However, some Weibo users discovered that the domestic AI grading system had failed.

As can be seen from the picture, this APP called Singularity Academy requires that the answer must be "completely consistent" with its formula when grading math problems. Even if the other steps are correct, full marks will not be given.

△ The picture comes from Weibo @Antolia

Even if the derivation process is correct and the answer is correct, only 7.5 points are given for a 10-point question.

So, do you believe that AI is more reliable than teachers in grading test papers?

Reference Links:

https://www.theverge.com/2020/9/2/21419012/edgenuity-online-class-ai-grading-keyword-mashing-students-school-cheating-algorithm-glitch

https://twitter.com/DanaJSimmons/status/1300997133311508480

https://www.reddit.com/r/edgenuity/comments/gbmx2l/how_do_i_successfully_cheat_on_edgenuity_without/

https://finance.yahoo.com/news/edgenuity-announces-recognitions-tech-learning-140000177.html