What is the difference between a student paying money to a human ghostwriter or prompting an AI if...
Most students today understand that modern plagiarism checkers are so good that copying/pasting from others, self-plagiarism, and collusion is not smart and can have consequences, as exemplified in The New York Times: “Norway’s higher education minister […] resigned after admitting she “took text from other assignments without stating the sources” in her master’s thesis.” [1].
Meanwhile, commercial ghostwriting, where no detection tool is available, has increased from a historical average of 3.5% of students to 15.7% reported in 2014 [2]. Then, with little forewarning, generative AI became available to everyone in late 2022, and research describes available AI detectors as “neither accurate nor reliable.” [3].
While the concern until now has been plagiarism, the quadrupling of human ghostwriting and the superpowered AI ghostwriter ChatGPT is shaking the cheating prevention landscape. In summary, cheating methods on essay-based assessment are Collusion, Plagiarism, Ghostwriting, and AI writing, with effective detection tools only existing for collusion and plagiarism.
Figure, Existing detection tools: Green is the cheating method the tool is designed to prevent, yellow is where the tool is unreliable, and red is where the tool is incapable.
Forbes: "89% of students admit to using ChatGPT for homework."
Before ChatGPT and COVID-19, 29.9% of students admitted to cheating on online exams, which jumped to a staggering 54.7% during COVID-19 [4]. While research on AI ghostwriting is still limited, an article in Forbes asserts that 89% of students admit to utilizing ChatGPT for homework-related purposes [5]. Homework is not a high-stakes assessment, but when new research suggests that “ChatGPT has the potential to blur the line between acceptable and unacceptable behavior from a student perspective.” [6], this should be a major concern for everyone involved in academic integrity. A concern The Sydney Morning Herald describes as: “Cheating students undermine their qualification and their university” [7].
Changing the whole way we assess is dramatic and AI detection is a flawed strategy
We should not feel defeated by AI, so returning to classroom exams is necessary. We should rather explore new technologies to verify original student writing so we don't have to force major changes to the assessment process. The key to assessing a student's ability is to assess original work. Changing how students are being assessed should evolve from a pedagogic discussion, not as a hasty response to a new threat to academic integrity.
An effective strategy against plagiarism has proven to be plagiarism detectors. But how do you detect ghostwriting and AI writing? In this blog post, you can read why AI detection is a flawed strategy (nor.education). Detection is not the way forward - we need to redefine the solution!
If we solve the age-old problem of ghostwriting, we have also solved AI writing
What is the difference between paying money to a human ghostwriter or prompting an AI for text to use in an essay? It doesn’t matter if the ghostwriter is human or AI. What matters is that the text is not original student work, hence a violation of universal academic integrity principles. In other words, AI writing will be solved if we solve the age-old ghostwriting problem.
If you consider AI writing a new form of ghostwriting, what would you do if you suspected ghostwriting? Compare the essay with previous writing from the student and have a conversation with the student about the content of the essay! This is an age-old and effective method to verify that the student's work is authentic, so how can it be automized and scaled in the same way as plagiarism detection?
The solution is to ensure Authenticity - “You detector”
Nor.Education has developed the Authorship Authentication System by combining well-established methods from authorship analysis research [8] and simulating a conversation with the student at the time of delivery.
Rather than looking for the ghostwriter (e.g., ChatGPT), Nor.Education validates the authenticity of the student by looking for "YOU”. An authorship verification model compares the newly submitted essay with historical and authentic student writing and judges the likelihood of the student being the authentic author of the essay. The text sample of authentic student writing can be acquired during matriculation or through an in-class writing session since as little as 200 words will be sufficient (more is better). New to the students is that they are requested to answer a series of questions about the content of their essay at the time of submission to collect further evidence of authorship.
The existing plagiarism checker will produce its usual text similarity report, the new authorship verificator will produce evidence-based conclusions from linguistics analysis, and a summary will show the correctness of the student's answers to the questions uniquely tailored to the essay. All in all, this will form a solid basis for measuring authenticity and automatically flagging high-risk submissions relative to the institution's acceptable threshold.
Authorship Authentication, in parallel with the plagiarism checker, safeguards against all cheating methods
When using the existing plagiarism checker combined with the new Authorship Authentication system, the solution covers all cheating methods, both the new problem with AI writing and the age-old Ghostwriting problem.
While it might not add material value, including an AI detector is not harmful if you understand its limitations. For example, as of March 2024, 6 out of the top 10 plugins in the “Writing” section on OpenAI’s Custom GPT Marketplace[9] are designed to evade AI detection.
In this blog post, you can learn about Authorship Authentication: A reliable way to verify student writing and combat ChatGPT/AI writing: Authorship Authentication (nor.education).
Figure, New cheating prevention tools: Green is the cheating method the tool is designed to prevent, yellow is where the tool is unreliable, and red is where the tool is incapable.
External References:
[1] The New York Times. A Norwegian Official Fought Plagiarism. Then She Was Caught Copying.https://www.nytimes.com/2024/01/25/world/europe/norway-plagiarism-minister-resigns.html
[2] Philip M. Newton, Swansea University Medical School (2018). How Common Is Commercial Contract Cheating in Higher Education and Is It Increasing? A Systematic Review. *https://www.frontiersin.org/articles/10.3389/feduc.2018.00067/full#B43*
[3] Debora Weber-Wulff (University of Applied Sciences HTW Berlin, Germany), (corresponding author) Alla Anohina-Naumeca (Riga Technical University) Sonja Bjelobaba (Uppsala University, Sweden), Tomáš Foltýnek (Masaryk University) Jean Guerrero-Dib (Universidad de Monterrey, Mexico), Olumide Popoola (Queen Mary University of London, UK), Petr Šigut (Masaryk University, Czechia), Lorna Waddington (University of Leeds, UK). (June 2023). Testing of Detection Tools for AI-Generated Text. *https://arxiv.org/abs/2306.15666*
[4] Philip M. Newton, Keioni Essex (2023) How Common is Cheating in Online Exams and did it Increase During the COVID-19 Pandemic? A Systematic Review https://link.springer.com/article/10.1007/s10805-023-09485-5#citeas
[5] Forbes (2023). Educators Battle Plagiarism As 89% Of Students Admit To Using OpenAI’s ChatGPT For Homework. https://www.forbes.com/sites/chriswestfall/2023/01/28/educators-battle-plagiarism-as-89-of-students-admit-to-using-open-ais-chatgpt-for-homework/
[6] Bergström, J., Repo, V., & Tuunainen, V. (2024). Challenges to Academic Integrity from New Tools-A Survey of Students' Perceptions and Behaviors of Employing ChatGPT.https://scholarspace.manoa.hawaii.edu/server/api/core/bitstreams/a07e09f2-dc38-4bf0-8094-f478df8f0c48/content
[7] The Sydney Morning Herald (2023) Cheating students undermine their qualification and their university. https://www.smh.com.au/national/nsw/cheating-students-undermine-their-qualification-and-their-university-20230917-p5e5a6.html
[8] PAN is a series of scientific events and shared tasks on digital text forensics and stylometry https://pan.webis.de/
[9] March 2024: 6 out of the top 10 plugins in the “Writing” section on OpenAI’s Custom GPT Marketplace are designed to evade AI detection.
