Balancing AI Use and Plagiarism in Education

TL;DR ChatGPT offers great benefits to the education industry but also raises concerns about increased plagiarism. AI-generated content can misrepresent students' knowledge, and detecting AI plagiarism is more challenging than traditional plagiarism. Educators are divided on how to approach the use of AI in homework and exams. Some have gone back to traditional pen-and-paper exams, while others are completely okay with their students using it. A better and more balanced approach maybe for instructors to allow its utilization to some extent, while monitoring the degree of originality in the students' writing.

Generative AI’s rapid rise 

Since the invention of the iPhone, and maybe the Internet, we haven’t witnessed technology that is as revolutionary as Generative AI. Generative AI will change or at least have an impact on every single industry in some shape or form. In fact, ChatGPT is so useful that it reached 1 million users in 5 days, making it the fastest growing consumer application in history.

Generative AI will revolutionize the world of education due to the vast array of benefits it offers to students and educators alike. By providing instantaneous, personalized assistance, Generative AI fosters an interactive and engaging learning experience that transcends geographical boundaries. For example you can ask ChatGPT to describe the most sophisticated concept, like Quantum Theory, but ask it to “Describe it as if you would to a 12 years old” or even ask it to “explain it as an Eminem rap song” and it does this astonishingly well. As a result, Generative AI empowers students to grasp complex concepts more effectively by bridging gaps in understanding, and provides access to high-quality educational resources irrespective of their socio-economic background. 

The problem

With any revolutionary invention, comes risks and threats that need to be mitigated. For example when the Internet was invented, security and cyber-security to be more precise became a major concern that needed continued investment and innovation to keep the Internet safe. Generative AI originality and plagiarism is going to be a fast growing concern. Generative AI enables individuals to present themselves as experts in any domain, sharing content on highly advanced topics and ideas. However, unlike in previous times when content originated from the reader's knowledge and experience, this information is now wholly generated by artificial intelligence.

This is going to be a problem in several domains for example there is a chance that in a few years many (if not most) of blogs or articles posted on the Internet will be almost completely created with generative AI to bring SEO or organic traffic via Search results to a company’s page, while this may not seem like a problem at first, the issue will be that likely no attribution will be made to the Chatbots that created the article. For example a recent article found that 

“CNET's AI Journalist Appears to Have Committed Extensive Plagiarism - CNET's AI-written articles aren't just riddled with errors. They also appear to be substantially plagiarized.”

However this problem is currently being felt the most in academic institutions. For example, we talked with a professor from a major university and one interesting quote we heard was that

“for some reason all the admission essays were very high quality this year.”

‍Also, an article recent published found that

“89% Of Students Admit To Using OpenAI’s ChatGPT For Homework”

To understand the magnitude of this issue in academia, it is crucial to examine the educational process students undergo. In primary and secondary education, educators must teach students and assess their understanding of specific subjects. A key aspect of this involves teaching students how to write without AI assistance, as writing is intrinsically linked to critical thinking.

In higher education, institutions aim to admit the most exceptional students and prepare them for life after graduation. Educators' accurate evaluation of students' subject comprehension enables them to assess their teaching techniques and modify their methods to address students' needs more effectively. Finally, companies seek to hire top-performing students, often using GPAs as a primary indicator of knowledge and academic prowess.

However, if students' writing is increasingly produced by AI, GPAs and essays (such as admission essays) will reflect the proficiency of an individual's usage of tools like ChatGPT, rather than their true understanding of a particular domain or subject. Consequently, this will provide a misleading representation of a student's knowledge and capabilities to schools, teachers, companies, and the students themselves.

To understand how powerful Generative AI and in particular GPT-4 is in giving answers to students, one only needs to look at the GPT-4 technical report provided by OpenAI. 

GPT-4 can provide a 90% percentile or better on most subject or exams

This isn’t an entirely new problem

Plagiarism isn’t a new problem. In fact, it became a significant concern with the advent of the Internet, as it provided students with easy access to a vast array of information and resources. While this access to knowledge was beneficial in many ways, it also made it easier for students to copy and present others' work as their own. Instructors were alarmed by this growing issue, as it not only threatened academic integrity but was also seen as a hindrance to students' development of original thinking and writing skills.

To combat plagiarism, various tools were developed that could cross-check students' work against existing content on the Internet. These plagiarism detection software helped identify instances of copied content, ensuring that academic standards were upheld. As a result, the problem of plagiarism was, to a some extent, considered resolved. 

It’s very different this time

While it’s easy to say this problem is similar to what we’ve seen in the past, i.e. invention of the internet or invention of the calculator, a quick comparison of traditional plagiarism vs AI plagiarism shows that it’s very different this time. A major difference is that Generative AI can answer almost any question with a unique and sophisticated answer and this answer can't be cross checked with text that previously existed on the Interent. A student only needs to put the question of an essay or exam in ChatGPT and copy the answer. To show how traditional plagiarism compares with AI plagiarism see the table below.

Two different approaches to the problem

The introduction of ChatGPT has led to varied perspectives among instructors regarding its use in homework and exams. Some educators are comfortable with students utilizing ChatGPT, with a mindset that schools and curriculums need to adapt to any new technology just as math class did with the invention of the calculator.

On the other hand, certain instructors are concerned that the use of ChatGPT in homework and exams could compromise the development of critical thinking and original ideas among students. In order to maintain academic integrity and ensure that students are not relying on AI-generated content for their assignments and tests, these educators have implemented various measures to block use of AI. These include traditional pen-and-paper exams or employing browser extensions that record students' faces while they type, providing evidence that the work is produced without AI assistance. 

A third approach - a more balanced one

An alternative approach that instructors can adopt in response to the use of generative AI in student assignments is to allow its utilization to some extent, while monitoring the degree of originality in the students' writing. This method can be compared to banks issuing credit cards based on customers' credit scores as a measure of their creditworthiness. While some banks may only issue credit cards to customers with high FICO scores (740+), others might be more lenient, catering to those with lower scores (580+). However, in both cases, the banks need to be aware of the customers' creditworthiness.

Similarly, in the context of academic writing, instructors can decide on the acceptable degree of originality for their students' work. Some instructors may be willing to accept essays with a lower originality score, while others may insist on a higher levels of uniqueness in the submitted assignments. Regardless of the threshold each instructor sets, they would still want to know the degree of originality present in the students' writing. This approach allows for flexibility while still maintaining an emphasis on the importance of original thought and creativity in academic work. By doing so, educators can strike a balance between leveraging technological advancements and upholding academic integrity.

With Rumi this is exactly our intention. Rumi provides originality scores, along with originality score breakdowns and all revisions of a document for further analysis if needed. 

‍