Friedrich, Ella and Rastegar, Sophia
Copy Catcher: Building an Advanced Academic Content
Authentication System
Introduction
The idea of AI as something that not only creates new things but builds upon man made
creations as well is a concept that has been around for a while. Popular programs such as
ChatGPT help with anything from writing essays and doing math homework, to being able to
code in multiple different languages. However as AI advances, so do methods of exploiting it.
According to a study conducted by the Pew Research Center in 2023, 1 in 5 students are using
AI tools like ChatGPT to do their assignments for them. These numbers, as stated by
EducationWeek, are only growing, and while AI detectors can be used for catching those who
rely on this cheating method, it is never 100% accurate. Multiple studies have been conducted on
the accuracy of these detection services, notably one by Ahmed M. Elkhatat, Khaled Elsaid and
Saeed Almeer, which stated that as the AI models have increased in their intelligence. Bashay
explains that the upgrade from ChatGPT 3.5 to its successor ChatGPT 4.0 showed a 7% decline
in the detectors capacity to do their intended purpose. This creates more of a margin for students
to cheat. The surge of false reports saying that students have used AI when they haven’t also
causes reasonable doubt in scenarios where it is assumed AI was used but cannot be proven due
to the amount of false positives. Looking at this information, the desired outcome of this
experiment is a new program whose software is able to detect both AI and non-AI cheating in
assignments through machine learning.
1
Friedrich, Ella and Rastegar, Sophia
The Coding Portion
The primary task involved in this project is training the AI to generate text that closely
mirrors the tone, structure and vocabulary of the provided sample of text. To achieve this we
utilize Python, a high-level programming language particularly well-suited for natural language
(NPL) and machine learning applications. Python's dominance in the field stems from its robust
ecosystem of libraries, including Hugging Face’s Transformers—a powerful toolkit for natural
language processing that provides easy access to pre-trained models like BERT and GPT—and
spaCy, which is widely used for tasks such as named entity recognition, part-of-speech tagging,
and syntactic parsing. Together, these tools offer advanced capabilities for text analysis and
model training, making Python the preferred language for many machine learning and NLP
applications. These libraries enable the AI to perform tasks, such as tokenization syntactic
Analysis and pre-train language models, which are crucial for mimicking writing styles.
(Columbia).
What distinguishes Python from other programming languages in this context is its deep
integration of powerful machine learning frameworks like TensorFlow and PyTorch, which
facilitates the development of complex neural networks (Sharma 2022). These frameworks along
with Python’s ability to handle fast datasets and perform computationally intensive tasks with
relative efficiency, make it an ideal choice for training sophisticated models that can learn
intricate patterns in language. By leveraging these tools, Python provides the flexibility and
computational power to process large volumes of text and refine AI’s ability to generate
coherent, contextual, appropriate essays that reflect nuisance specific writing styles.
2
Friedrich, Ella and Rastegar, Sophia
AI Tracking System
AI trackers designed to detect and distinguish between human and AI generated content
typically rely on advanced machine learning models that analyze various features of the data.
Sharma said these systems use algorithms to evaluate patterns in the content, such as linguistic
anomalies, inconsistencies in writing style, and unnatural sentence structures that are indicated
by machine generated text. In the case of tracking AI content on platforms like [Link] for
instant AI detection systems, assess factors like syntax, grammar, coherence, and semantic
patterns, comparing the input text to known AI writing models. The tracker identifies specific
markers, such as repetitive phrases, overuse of certain word choices, or lack of announced
human-like error patterns that are common in text produced by AI. These systems, often employ
specialized models trained on large data sets containing both human and AI generated text to
classify flag content as potentially artificial by leveraging, statistical and computational
techniques, AI trackers can distinguish between human and machine authorship with increasing
accuracy, providing an important tool, for Content verification and academic integrity.
Purpose
Students often use AI or even pay others to complete their assignments instead of doing the work
themselves. The purpose of this experiment is to create a different AI program type that detects
cheating in assignments.
Work Cited
3
Friedrich, Ella and Rastegar, Sophia
Beshay. (2024, April 14). About 1 in 5 U.S. teens who’ve heard of ChatGPT have used it for
schoolwork. Pew Research Center.
[Link]
of-chatgpt-have-used-it-for-schoolwork/
Prothero, A. (2024, October 24). New data reveal how many students are using AI to cheat.
Education Week.
[Link]
cheat/2024/04
Anoriega. (2024, October 1). Python and Its Basics: A Beginner’s Guide | Columbia Engineering
Boot Camps. Columbia Engineering Boot Camps.
[Link]
Sharma, G. (2022, December 20). Comparing Python to other popular programming languages.
Medium.
[Link]
15fb757aeaa1
The Python tutorial. (n.d.). Python Documentation.
[Link]
4
Friedrich, Ella and Rastegar, Sophia