Academia and Industry Synergy: Addressing Integrity Challenge in

Programming Education

Rina Azoulay

, Tirza Hirst

and Shulamit Reches

Department of Computer Science, Jerusalem College of Technology, Jerusalem, Israel

Department of Mathematics, Jerusalem College of Technology, Jerusalem, Israel

Keywords:

ChatGPT, Large Language Models, Computer Science Education, Plagiarism, Integrity, LLMs.

Abstract:

This research addresses the profound challenges presented by sophisticated large language models (LLMs)

like ChatGPT, especially in the context of educational settings, focusing on computer science and program-

ming instruction. State of the art LLMs are capable of generating solutions for standard exercises that are

assigned to students to bolster their analytical and programming skills. However, the ease of using AI to

generate programming solutions poses a risk to the educational process and skill development, as it may lead

students to depend on these solutions instead of engaging in their own problem-solving efforts. Our study

suggests collaborative methods involving computer science educators and AI developers to provide evaluators

with tools to distinguish between code produced by ChatGPT and code genuinely created by students. We

propose a novel steganography-based technique for watermarking AI-generated code. By implementing this

comprehensive strategy and effectively utilizing such technology through the combined efforts of educators,

course administrators, and partnerships with AI developers, we believe it is possible to preserve the integrity

of programming education in an age increasingly inﬂuenced by LLMs capable of generating code.

1 INTRODUCTION

Large language models (LLMs), such as ChatGPT,

can be a valuable resource for programmers, espe-

cially those looking for assistance, inspiration, or

seeking clariﬁcation on programming concepts. They

are trained on a vast amount of programming-related

text, which enables them to understand and produce

code in various programming languages. They can

assist with tasks such as identifying syntax errors,

suggesting code corrections, or explaining language-

speciﬁc features.

As a result, LLMs’ ability to perform high-level

cognitive tasks and produce human-like text has

raised concerns about their potential role in academic

dishonesty (Susnjak, 2022).

One of the signiﬁcant challenges that has emerged

as a result of LLMs is their ability to provide solu-

tions for programming tasks that students tradition-

ally tackle to sharpen their analytical and writing

skills. In particular, ChatGPT and other tools can as-

sist in the software development process, from code

generation to optimization (Azaria et al., 2023).

While these tools offer assistance in software de-

velopment, from coding to optimization, their use in

educational settings can be problematic. In program-

ming education, the aim of programming tasks is to

enable students to practice problem-solving and pro-

gramming, to enhance their analytical and program-

ming skills. Given LLMs’ ability to produce code

solutions, there is a considerable temptation for stu-

dents to utilize these ready-made solutions instead of

engaging in self-practice and skill development.

Relying on AI-generated solutions could hinder

the student’s learning process and obstruct the cru-

cial development of their skills. Furthermore, LLMs

can produce erroneous or unsuitable outputs, a phe-

nomenon known as ’hallucination’. Reliance on these

tools can lead to students absorbing incorrect infor-

mation, developing a ﬂawed understanding of the sub-

ject matter, and adopting approaches unsuited to the

task at hand, owing to this propensity of LLMs.

The challenge of ensuring integrity in academic

assignments given the current AI abilities occupies

the best teachers and lecturers around the world.

In this paper, we propose a comprehensive set of

strategies, designed to ensure integrity in program-

ming tasks. To address plagiarism concerns, we sug-

gest a collaborative method involving educators, as-

signment creators, and AI companies, focused on

Azoulay, R., Hirst, T. and Reches, S.

Academia and Industry Synergy: Addressing Integrity Challenge in Programming Education.

DOI: 10.5220/0012451000003636

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 16th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2024) - Volume 3, pages 1135-1143

ISBN: 978-989-758-680-4; ISSN: 2184-433X

1135

a steganography-based method (Singh et al., 2009).

A steganography-based method refers to a technique

where information is hidden within another medium

to prevent its detection. In the context of AI-

generated code, we suggest embedding hidden water-

marks within the text of the code to clearly indicate its

AI origin. These concealed markers enable teachers

and instructors to swiftly detect plagiarism, thereby

ensuring accountability by visibly attributing the AI-

created content to its source.

By applying this method, we aspire to construct a

balanced and sustainable ecosystem of AI use within

the educational sphere, effectively distinguishing be-

tween AI-generated snippets and student-crafted code

to ﬂag instances of academic dishonesty.

A major point for deliberation is what might mo-

tivate the developers of AI tools to cooperate in iden-

tifying code fragments produced by their systems.

Nevertheless, we advocate establishing legal bound-

aries and obligations that require AI companies to ad-

here to speciﬁc criteria, ensuring responsible and eth-

ical AI use in the educational domain. With the global

momentum towards crafting regulations for the ethi-

cal and safe application of AI, it is reasonable to antic-

ipate the emergence of such rules in this ﬁeld. These

regulations could speciﬁcally include offering special

licenses to lecturers and teachers, allowing them to

use the AI tool and request the implementation of

steganography-based techniques. Under these regu-

latory measures, any company entering this market

would need to adhere to these stipulations and em-

bed identiﬁable markers in their code. Additionally,

within the context of group licenses for academic in-

stitutions using commercial LLM tools, these tools

would be required to comply with these regulatory

standards. This requirement could incentivize com-

mercial entities to meet these standards and integrate

watermarks into their code, driven by ﬁnancial mo-

tives.

In our study, we explore existing literature on

text-based steganography. We then examine differ-

ent methods for integrating these techniques into AI-

generated code. Additionally, we propose communi-

cation protocols that facilitate collaboration between

teachers and AI systems, aiming to produce the re-

quired encoded outputs.

The remainder of this paper is organized as fol-

lows: Section 2 presents a literature review on aca-

demic integrity in the context of AI tools. Section 3

provides an overview of text steganography and its

use in identifying AI-generated content to prevent pla-

giarism. Section 4 outlines various steganographic

methods for embedding unique markers in automat-

ically generated code. Section 5 describes a collabo-

rative approach between educators and AI developers

to utilize steganography for detecting AI-generated

content. Finally, the paper wraps up with Section 6,

where we present our conclusions and suggest areas

for further study.

2 RELATED WORK

We proceed by offering a detailed overview of vari-

ous strategies and recommendations aimed at uphold-

ing academic integrity, particularly in light of the ca-

pability of LLMs to produce intricate content. When

ChatGPT was unveiled as the inaugural LLM for pub-

lic use, it spurred widespread discourse both among

the general public and speciﬁcally within educational

institutions. The dialogue often centered on its appro-

priate utilization and its broader societal impact.

The integration of ChatGPT into student educa-

tion brings with it a multitude of opportunities and

challenges. Neumann et al. (Neumann et al., 2023)

focuses on articles addressing the impact of Chat-

GPT on higher education, speciﬁcally in the areas

of software engineering and scientiﬁc writing. The

paper recommends utilizing plagiarism checkers and

AI detection tools, or manually examining the texts

for ChatGPT ﬁngerprints. Additionally, it suggests

implementing an oral examination or requiring doc-

umentation of the examination process. Similarly,

the review (Lo, 2023) on the impact of ChatGPT’s

on education, reveals the capabilities of it across sub-

jects (strong in economics, moderate in programming,

weak in math), and suggests that schools adopt assess-

ment methods, update policies, train instructors, and

educate students to effectively integrate ChatGPT.

Excessive reliance on AI may have lasting impacts

and potentially undermine the professional develop-

ment of upcoming generations. The study referenced

in (Mijwil et al., 2023) discusses concerns about the

next generation relying solely on AI to complete tasks

without putting in effort, while emphasizing the im-

portance of educating them about the limitations and

potential biases of AI and how to evaluate the infor-

mation it provides. The paper’s main conclusion is

that artiﬁcial intelligence applications like ChatGPT

function as tools to support human work rather than

replacing it entirely. While they can assist in task

completion and enhance the quality of writing, they

cannot completely replace human expertise in writing

and critical thinking.

As highlighted in our introduction, the impact

on computer science education holds unique signif-

icance. This topic is further explored by Quersi

(Qureshi, 2023), who delves into the integration of

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1136

ChatGPT in undergraduate computer science curric-

ula, speciﬁcally focusing on foundational program-

ming concepts. An experiment was organized with

two distinct groups of students, divided into teams,

tasked with solving programming assignments: one

group had access only to textbooks and notes with

no internet, while the other group was equipped with

ChatGPT. As it turned out, the teams using Chat-

GPT achieved higher average grades compared to

the group that did not use it across all programming

tasks. Nonetheless, they also spent more time grap-

pling with the most complex problems. Finally, ad-

vantages and disadvantages of ChatGPT in teaching

computer science are discussed, and various strategies

are suggested to avoid misuse of ChatGPT by pro-

gramming students, including the use of automated

tools to detect plagiarism.

However, detecting AI-generated content remains

a formidable challenge. While numerous tools have

been developed, their efﬁcacy is yet to be fully real-

ized. In a study conducted by Khalil et al. (Khalil

and Er, 2023), the aim was to explore the originality

of content generated by ChatGPT. To accomplish this,

the researchers employed two popular plagiarism de-

tection tools to assess the originality of 50 essays pro-

duced by ChatGPT on various topics. The ﬁndings of

the study manifest that ChatGPT possesses a signif-

icant potential to generate sophisticated and intricate

text outputs, largely eluding detection by plagiarism-

checking software.

To address complexities involved in identifying

and mitigating academic dishonesty, Cotton et al.

(Cotton et al., 2023) propose the establishment of

policies and procedures, provision of training and

support, and utilization of diverse methods to detect

and prevent cheating. In addition, they suggest that

teachers may consider mandating a written declara-

tion from students asserting the originality of their

work. However, such a declaration may not be gen-

uine in actual situations. The question is what can

be done to encourage students to submit independent

work.

Considering foreign language studies, Perkins

(Perkins, 2023) highlights the use of LLMs by for-

eign Language (EFL) learners, including potential as-

sistance in digital writing and beyond, and delves into

the concerns regarding academic integrity associated

with students’ use of these tools. He suggests that

the use of LLMs should not be considered plagiarism,

provided that students clearly disclose their use of the

technology in their submissions. Moreover, there are

legitimate uses of these tools in student education. He

concludes that the determination of whether a speciﬁc

use of LLMs by students is deemed academic miscon-

duct should be based on the academic integrity poli-

cies of the institution. These policies must be updated

to reﬂect how these tools will be utilized in upcoming

educational settings.

We proceed by describing some studies suggest-

ing how to incorporate the appropriate use of LLMs

in education while handling its challenges and limi-

tation. Kasneci et al. (Kasneci et al., 2023) caution

against over-reliance, emphasizing the importance of

recognizing LLM limitations and promoting teacher

training. The authors advocate the use of LLMs as

supplementary tools alongside other educational re-

sources, fostering student creativity through indepen-

dent projects, and integrating critical thinking into

curricula. They also stress the signiﬁcance of human

oversight in reviewing LLM outputs and the neces-

sity of a strategy focused on critical thinking and fact-

checking for effective LLM integration.

Kumar et al. (Kumar et al., 2022) detail how LLM

technology targets education and propose recommen-

dations that educators may adopt to ensure academic

integrity in a world with pervasive LLM tools. They

emphasize the importance of fostering a genuine de-

sire to learn and develop deep research skills among

students to counteract the effects of LLM generators.

They recommend to prioritize conferencing with stu-

dents about their writing to foster collaboration and

skill development, although this approach is challeng-

ing in large size classes.

We end this section with a description of two stud-

ies related to the impact of LLMs on higher edu-

cation. Tlili et al. (Tlili et al., 2023) conducted a

qualitative study to explore the impact of ChatGPT

on education, structured in three phases. In the ﬁrst

phase, a social network analysis of tweets showed

that the majority of public sentiment on social me-

dia was positive towards ChatGPT, with positive sen-

timents were expressed nearly twice as much as neg-

ative ones. The second phase involved interviews

with 19 stakeholders who blogged about their experi-

ences with ChatGPT. The analysis revealed that many

viewed ChatGPT as transformative for education, but

there were concerns about its potential to hinder in-

novation and critical thinking. While many found

ChatGPT’s responses satisfactory, some noted occa-

sional errors and outdated information. Ethical con-

cerns included potential plagiarism, misinformation

risks, and privacy issues. The ﬁnal phase presented

user experiences from ten educational scenarios with

ChatGPT, highlighting challenges such as cheating,

truthfulness, and potential manipulation.

Sullivan et al. (Sullivan et al., 2023) analyzed 100

news articles to study ChatGPT’s inﬂuence on higher

education across four countries (US, UK, Australia

Academia and Industry Synergy: Addressing Integrity Challenge in Programming Education

1137

and NZ), published between 2020 and February 2023.

Key themes included academic integrity concerns,

potential AI misuse, and debates on cheating. In re-

sponse, some universities reintroduced supervised ex-

ams, while others focused on assignments demanding

critical thinking. Institutional policies varied, with

some banning ChatGPT and others permitting its use

under conditions, often citing its inevitable workplace

integration. Many articles also provided strategies to

combat plagiarism and promote student originality.

The numerous educational beneﬁts of ChatGPT

include personalizing learning experiences and en-

hancing employability as AI reshapes industries. It

aids non-traditional and non-native English-speaking

students, serves as a quasi-translator, and provides

tools for those with disabilities, mitigating associated

stigmas. However, concerns regarding ChatGPT in-

clude the potential for spreading disinformation and

producing inaccurate information (’hallucinations’),

as well as issues related to copyright infringement,

privacy violations, and data security. While some

worry about students losing critical thinking abilities,

others champion the integration of AI into teaching

and assignments.

Finally, the authors warn that portraying ChatGPT

mainly as a cheating tool rather than a learning aid

can shape public opinions about university education,

inﬂuence academic reactions, and affect student per-

spectives on the appropriate use of this tool. Stu-

dents exposed to articles about cheating with Chat-

GPT might be more inclined to cheat themselves. Re-

search shows that the perception of frequent cheating

opportunities increases the actual incidence of cheat-

ing among students.

Our study emphasizes achieving integrity within

computer science education and outlines active strate-

gies to enhance self-work on programming exercises.

In particular, we detail practical implementations and

suggestions to enhance integrity, focusing on educa-

tion methods, as well as plagiarism detecting meth-

ods, used by course teams with or without the coop-

eration of AI developers.

3 TEXT STEGANOGRAPHY

We begin with an overview of text steganography,

followed by suggestions on how it can be employed

to identify AI-generated content, thereby preventing

various forms of plagiarism.

Text steganography refers to methods of using text

as a means to conceal information (Singh et al., 2009).

In our context, we need this type of steganography,

since code is represented as text, without any way to

change letter sizes, fonts, etc. We can only change the

characters themselves. Text steganography is notably

more challenging than other forms of steganography,

primarily due to the minimal redundant data available

in a text ﬁle as opposed to multimedia ﬁles such as

images or audio.

Numerous studies, such as (Singh et al., 2009; Por

et al., 2012; Dulera et al., 2012; Roy and Manasmita,

2011; Bender et al., 1996; Agarwal, 2013; Hariri

et al., 2011; Delina, 2008; Shirali-Shahreza, 2008),

focus on text steganography and encryption methods.

Krishnan et al. (Krishnan et al., 2017) provide a com-

prehensive review and classiﬁcation of text steganog-

raphy techniques, along with a comparison of existing

approaches.

Additionally, some studies suggest using ad-

vanced algorithms and machine learning methods for

the text steganography task (Xiang et al., 2020; Satir

and Isik, 2012; Satir and Isik, 2014; Fang et al., 2017;

Yang et al., 2020; Bhattacharyya et al., 2009). In

general, the approach to text steganography relies on

utilizing the unique properties of text ﬁles. These

unique properties provide opportunities to hide infor-

mation. For instance, one could subtly alter the text

document’s structure (in our case, the program code

and comments) to incorporate concealed information

while ensuring the changes are subtle enough not to

arouse suspicion or signiﬁcantly alter the output.

Additional strategies could involve the use of

typographical errors, the positioning of spaces, or

even the application of invisible characters. Another

method could be the crafting of sentences that hold

dual meanings. Here, the literal interpretation main-

tains the appearance of a standard document, while

dedicated software will be able to reveal the informa-

tion encoded in the text.

4 UTILIZING

STEGANOGRAPHIC

METHODS FOR

FINGERPRINTED CODE

Our focus is on developing a strategy that allows

ChatGPT to insert unique markers or ”digital ﬁnger-

prints” into its generated code snippets. Stegano-

graphic methods that manipulate white space to con-

ceal messages (Bender et al., 1996; Por et al., 2008)

appear to be suitable for this application. However,

techniques like line-shift and word-shift (Roy and

Manasmita, 2011; Hariri et al., 2011), which necessi-

tate altering the spatial layout of text, are not feasible

here. We will now examine multiple techniques that

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1138

ChatGPT can utilize to insert hidden messages into its

output code:

• White Spaces Encoding. Using a combination

of ”space” and ”tab” characters within blank lines

can facilitate binary encoding (Por et al., 2008).

This approach allows both data and control mes-

sages to be encoded by adjusting the distribution

of extra spaces and tabs. For instance, one might

designate ”space” to represent 0 and ”tab” to rep-

resent 1. This binary representation can convey

pertinent information about the text, such as its

date or the subject matter. The encoded infor-

mation can then be strategically placed within the

text, such as at the ends of lines or within empty

lines separating code sections. Figure 3 demon-

strates how a ”GPT” ﬁngerprint can be embed-

ded within the generated code using only invisi-

ble tabs and spaces, which can be viewed by se-

lecting the entire code within the Colab frame-

work. Figure 2 depicts a code generated by Chat-

GPT (OpenAI, 2023) that aims to produce such

a tabs-and-spaces string when given input text to

encrypt. Though students can detect and remove

these whites pace strings, they can be dispersed

throughout the code, making them hard to iden-

tify and eliminate.

• Steganography Using Different Types of

Comments. For a language that includes two or

more comment environments, for example, ”//”

or ”/* */” in C++, or ”#” and extended string in

Python, a coding can be performed by choosing

different types of comments and using them as

a signaling method, in a way determined by the

teacher and the AI engine.

• Unique Identiﬁers for Variables and Functions.

Various techniques can be employed to craft dis-

tinctive names for identiﬁers, such as variables,

functions, classes, etc., enabling the detection of

the code’s origin. Some examples include: us-

ing a speciﬁc and rare name, for some auxiliary

variable, utilizing a mix of uppercase and lower-

case letters at the start of an identiﬁer to embed

speciﬁc information; alternating between rounded

or angular letters in English (Dulera et al., 2012);

using identiﬁers with a certain common denom-

inator, for example, with a speciﬁc sum of their

ASCII values, or with a speciﬁc modulo value,

when dividing by a certain k, that will be chosen

in advance. Another possibility is selecting identi-

ﬁer names such that a hash function applied to the

name yields a particular value or possesses a cer-

tain characteristic (like being even, prime, etc.).

It is also viable to use the names of variables that

have a certain common denominator, for example

that the sum of the ASCII values of all the charac-

ters in the names of these variables will be 0 mod-

ulo k, for a certain k that the lecturer will choose

in advance.

• Insert Lines of Code and Comments with

Certain Characteristics. In addition to the afore-

mentioned methods, one can encrypt a text that

details the exercise’s origin and embed this en-

crypted version within the code comments, or

implant a speciﬁc message inside the original

comment, with certain speciﬁc typos (Shirali-

Shahreza, 2008), etc. While it may look like a

linguistic error at ﬁrst glance, it would indeed rep-

resent a concealed message. Broadly speaking,

various text steganography techniques – like syn-

onyms, linguistic mistakes, or words that have dif-

ferent spellings in British and American English –

can be integrated into the comment sections of the

generated code.

• Clear and Visible Markings. Beyond the

steganography-based solutions discussed, we can

also embed clear and unequivocal markers indi-

cating the code’s origin and purpose. This could

take the form of a header comment, along with in-

ternal comments, specifying that the subsequent

code was generated by the AI tool. Deleting these

identiﬁers would demand considerable editing on

the part of the student, potentially deterring pla-

giarism. An example of detecting work submitted

with such a comment is illustrated in Figure 1.

It is recommended that the encryption be in local

ranges (and not spread over the entire document), so

that if the student changes something in the output

program, not immediately all the encryption will be

compromised. When the encryptions are local and in

different places along the code, it will require thor-

ough and even Sisyphean work on behalf of the stu-

dent to identify and delete the various watermarks.

Apparently, dedicated software can be created to per-

form this editing, and this is also an issue that the reg-

ulator will have to handle in an appropriate manner.

5 ENHANCING COOPERATION

BETWEEN TEACHING STAFF

AND AI COMPANIES

We proceed by describing how steganography meth-

ods can be applied and recognized by the education

staff. In this study, we propose three approaches:

The ﬁrst involves signs that can be independently pro-

duced by an AI tool and recognized by any teacher

Academia and Industry Synergy: Addressing Integrity Challenge in Programming Education

1139

Figure 1: Clear and visible marking: An illustration.

Figure 2: ChatGPT generated code to provide steganography signature.

Figure 3: White-space based steganography for a ﬁngerprint.

or student, with or without the aid of AI detection

software. The second option employs steganography

applied independently by an AI tool, which then re-

ports in a dedicated manner to a speciﬁc teacher, as-

suming cooperation between the teacher and the sys-

tem. Lastly, the third option allows the teacher to re-

quest the AI tool to produce speciﬁc encoded mes-

sages when given certain prompts or when asked to

provide a particular generated code.

How can the cooperation between an exercise

providers and an AI company be carried out?

First, we assume that lecturers and teachers will

be able to obtain a special license, that will enable

them to cooperate on this issue with the AI company.

Before giving an assignment to the students, the lec-

turer will inform the AI company that he is giving a

certain assignment, and ask it to implant a secret mes-

sage inside each code it issues at the request of some-

one who requests a code for this assignment, within

a certain time frame, for example - within the com-

ing week. According to the second approach, the AI

company itself will report to the lecturer the message

it encrypted in every assignment it received on a cer-

tain topic. If there are special licenses, the lecturer

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1140

can ask the AI company to report to him and even

send him any piece of code that he issued at the re-

quest of a student on a speciﬁc topic, within a certain

time frame and even from a certain geographical area.

All this within the framework of the law, subsequent

to the enactment of the appropriate laws in this ﬁeld.

According to the third approach, the lecturer will

be able to guide the AI company, on which technique

to use, and what messages to implant.

The following algorithm describes a com-

munication process for using such a method of

steganography with cooperation between the teacher

and the AI generator, to detect the contents it gener-

ated:

(1) The instructor creates a distinctive assignment

with speciﬁc instructions that can be easily identiﬁed.

For instance, the task might require printing the string

”BLABLA” when a particular condition is satisﬁed or

evaluating a unique formula using a speciﬁed number.

(2) The lecturer reports to the AI company about

the exercise details and how it can be recognized. The

exercise details may include some unique contents,

such as unique requests (for example, printing the

”BLABLA” string), the exercise date interval, and

geographical area.

(3) In the second approach, the AI engine deter-

mines the coding method to employ. Meanwhile, in

the third approach, the lecturer designates the speciﬁc

steganographic techniques that will be implemented.

At this stage, choices are made regarding the stegano-

graphic method, the text targeted for encoding, an

associated encoding algorithm, a predetermined

cipher for encryption, etc.

(4) When an appropriate prompt, to perform this

unique task, is detected by the AI engine (according

to the second or third approach), it will produce the

steganographic method determined earlier within the

generated code output. For example, when the AI

tool is asked to print the string ”BLABLA” given the

above particular condition.

(5) The lecturer will be able to use dedicated

software to identify code snippets that students have

copied from the AI tool, using the steganographic

signs implanted by the AI engine.

It is important to acknowledge that requesting the

AI tool to identify users who pose a speciﬁc ques-

tion could raise privacy concerns. Therefore, we pro-

pose a solution that does not necessitate such report-

ing. In summary, this section suggests a collabora-

tive approach between educators and AI developers

to tackle plagiarism issues exacerbated by AI tools

such as ChatGPT. By leveraging the techniques of

text steganography, the proposal is to embed unique

”ﬁngerprints” within the code generated by the AI

tool, thus allowing educators to identify AI-generated

submissions. This system can be enforced by legal

frameworks, ensuring that AI developers are moti-

vated to participate. The ultimate aim is to uphold

academic integrity amidst the challenges posed by

rapidly advancing technology.

6 CONCLUSIONS

In the rapidly evolving era of AI technology, we are

faced with both exhilarating opportunities and daunt-

ing challenges. AI tools offer remarkable educational

beneﬁts, including interactive learning environments,

personalized study modules, and extensive knowl-

edge resources, particularly in areas like computer

science and programming assistance. Yet, these ad-

vancements also bring forth signiﬁcant concerns, such

as the inclination of students towards academic short-

cuts or an excessive reliance on automated solutions,

which can impede their skill development.

In this study, we examined the complex relation-

ship between AI tools like ChatGPT and program-

ming education, underscoring their beneﬁts and chal-

lenges. We particularly focused on the risk of students

becoming overly dependent on these tools, potentially

impeding their skill development and threatening aca-

demic integrity.

Our proposed solution involves a collaborative

strategy between AI developers and educators, cen-

tered on integrating unique encrypted text markers

into AI-generated responses. These markers are

aimed at helping educators identify when students

have used AI assistance in their programming tasks.

Our approach is designed not only to uphold aca-

demic integrity but also to foster genuine skill devel-

opment among students. By clearly differentiating

between student-generated work and AI-generated

content, we aim to deepen student engagement with

their learning materials. This method encourages a

more authentic and meaningful educational experi-

ence, ensuring students truly beneﬁt from their stud-

ies.

Future research should concentrate on developing

sophisticated methods for identifying AI-generated

content, ensuring its responsible application in educa-

tion. This involves the creation and evaluation of al-

gorithms for embedding watermarks in AI responses.

Testing the effectiveness of encrypted text mark-

Academia and Industry Synergy: Addressing Integrity Challenge in Programming Education

1141

ers in real classroom settings is a key aspect of

this work, providing essential insights for the practi-

cal implementation of these methods in educational

environments. Furthermore, collaboration between

academia and the tech industry, supported by stan-

dardized guidelines and collaborative platforms, is vi-

tal for seamlessly integrating AI into educational sys-

tems.

It is also crucial to study the long-term effects on

students who use AI tools like ChatGPT. Understand-

ing the impact of these tools on learning outcomes,

student engagement, and skill development over time

is highly signiﬁcant.

REFERENCES

Agarwal, M. (2013). Text steganographic approaches: a

comparison. arXiv preprint arXiv:1302.2718.

Azaria, A., Azoulay, R., and Reches, S. (2023). Chatgpt

is a remarkable tool – for experts. arXiv preprint

arXiv:2306.03102.

Bender, W., Gruhl, D., Morimoto, N., and Lu, A. (1996).

Techniques for data hiding. IBM Systems Journal,

35(3.4):313–336.

Bhattacharyya, D., Das, P., Bandyopadhyay, S. K., and

Kim, T.-h. (2009). Text steganography: a novel ap-

proach. International Journal of Advanced Science

and Technology, 3(1).

Cotton, D. R., Cotton, P. A., and Shipway, J. R. (2023).

Chatting and cheating: Ensuring academic integrity

in the era of chatgpt. Innovations in Education and

Teaching International, pages 1–12.

Delina, B. (2008). Information hiding: A new approach

in text steganography. In Proceedings of the Inter-

national Conference on Applied Computer and Ap-

plied Computational Science, World Scientiﬁc and En-

gineering Academy and Society (WSEAS 2008), pages

689–695.

Dulera, S., Jinwala, D., and Dasgupta, A. (2012). Experi-

menting with the novel approaches in text steganogra-

phy. arXiv preprint arXiv:1203.3644.

Fang, T., Jaggi, M., and Argyraki, K. (2017). Generat-

ing steganographic text with lstms. arXiv preprint

arXiv:1705.10742.

Hariri, M., Karimi, R., and Nosrati, M. (2011). An introduc-

tion to steganography methods. World Applied Pro-

gramming, 1(3):191–195.

Kasneci, E., Seßler, K., K

uchemann, S., Bannert, M.,

Dementieva, D., Fischer, F., Gasser, U., Groh, G.,

unnemann, S., H

ullermeier, E., et al. (2023). Chat-

gpt for good? on opportunities and challenges of large

language models for education. Learning and Individ-

ual Differences, 103:102274.

Khalil, M. and Er, E. (2023). Will chatgpt get you caught?

rethinking of plagiarism detection.

Krishnan, R. B., Thandra, P. K., and Baba, M. S. (2017). An

overview of text steganography. In 2017 Fourth Inter-

national Conference on Signal Processing, Communi-

cation and Networking (ICSCN), pages 1–6. IEEE.

Kumar, R., Mindzak, M., Eaton, S. E., and Morrison, R.

(2022). Ai & ai: Exploring the contemporary intersec-

tions of artiﬁcial intelligence and academic integrity.

In Canadian Society for the Study of Higher Educa-

tion (CSSHE).

Lo, C. K. (2023). What is the impact of chatgpt on edu-

cation? a rapid review of the literature. Education

Sciences, 13(4).

Mijwil, M. M., Hiran, K. K., Doshi, R., Dadhich, M., Al-

Mistarehi, A.-H., and Bala, I. (2023). Chatgpt and

the future of academic integrity in the artiﬁcial intelli-

gence era: a new frontier. Al-Salam Journal for Engi-

neering and Technology, 2(2):116–127.

Neumann, M., Rauschenberger, M., and Sch

on, E.-M.

(2023). “we need to talk about chatgpt”: The future of

ai and higher education. In 2023 IEEE/ACM 5th Int.

Workshop on Software Engineering Education for the

Next Generation (SEENG), pages 29–32.

OpenAI (2023). Chatgpt conversation. https://chat.openai.

com/. Accessed: 2023-07-17.

Perkins, M. (2023). Academic integrity considerations of

ai large language models in the post-pandemic era:

Chatgpt and beyond. Journal of University Teaching

& Learning Practice, 20(2):07.

Por, L. Y., Ang, T., and Delina, B. (2008). Whitesteg: a new

scheme in information hiding using text steganogra-

phy. WSEAS transactions on computers, 7(6):735–

745.

Por, L. Y., Wong, K., and Chee, K. O. (2012). Unis-

pach: A text-based data hiding method using unicode

space characters. Journal of Systems and Software,

85(5):1075–1082.

Qureshi, B. (2023). Exploring the use of chatgpt as a tool

for learning and assessment in undergraduate com-

puter science curriculum: Opportunities and chal-

lenges. arXiv preprint arXiv:2304.11214.

Roy, S. and Manasmita, M. (2011). A novel approach to

format based text steganography. In proceedings of

the 2011 international conference on communication,

Computing & Security, pages 511–516.

Satir, E. and Isik, H. (2012). A compression-based text

steganography method. Journal of Systems and Soft-

ware, 85(10):2385–2394.

Satir, E. and Isik, H. (2014). A huffman compression based

text steganography method. Multimedia tools and ap-

plications, 70:2085–2110.

Shirali-Shahreza, M. (2008). Text steganography by chang-

ing words spelling. In 2008 10th International Confer-

ence on Advanced Communication Technology, vol-

ume 3, pages 1912–1913.

Singh, H., Singh, P. K., and Saroha, K. (2009). A sur-

vey on text based steganography. In Proceedings of

the 3rd National Conference, pages 332–335. Bharati

Vidyapeeth’s Institute of Computer Applications and

Management.

Sullivan, M., Kelly, A., and McLaughlan, P. (2023). Chat-

gpt in higher education: Considerations for academic

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1142

integrity and student learning. Journal of Applied

Learning & Teaching, 6(1):1–10.

Susnjak, T. (2022). Chatgpt: The end of online exam in-

tegrity? arXiv preprint arXiv:2212.09292.

Tlili, A., Shehata, B., Adarkwah, M. A., Bozkurt, A.,

Hickey, D. T., Huang, R., and Agyemang, B. (2023).

What if the devil is my guardian angel: Chatgpt as

a case study of using chatbots in education. Smart

Learning Environments, 10(1):15.

Xiang, L., Yang, S., Liu, Y., Li, Q., and Zhu, C. (2020).

Novel linguistic steganography based on character-

level text generation. Mathematics, 8(9):1558.

Yang, Z., Wei, N., Liu, Q., Huang, Y., and Zhang, Y. (2020).

Gan-tstega: Text steganography based on generative

adversarial networks. In Digital Forensics and Water-

marking: 18th International Workshop, IWDW 2019,

Chengdu, China, November 2–4, 2019, Revised Se-

lected Papers 18, pages 18–31. Springer.

Academia and Industry Synergy: Addressing Integrity Challenge in Programming Education

1143