💬 NLP

Use of AI Tools: Guidelines to Maintain Academic Integrity in Computing Colleges

This paper addresses the challenges and opportunities of integrating AI tools in computing education by classifying assessment types, proposing general and specific guidelines to balance pedagogical benefits with academic integrity, and introducing a formal mathematical model for evaluating student assessments in the presence of AI assistance.

Original authors: Hatem M. El-boghdadi, Toqeer Ali Syed, Ali Akarma, Qamar Wali

Published 2026-04-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Hatem M. El-boghdadi, Toqeer Ali Syed, Ali Akarma, Qamar Wali

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the world of education, especially in computer science, as a giant, bustling kitchen. For years, chefs (students) have been learning to cook by chopping their own vegetables, measuring their own spices, and stirring their own pots. This is how they learn the craft.

Suddenly, a magical, all-knowing sous-chef named AI (like ChatGPT) has appeared. This robot can chop vegetables in a second, write perfect recipes, and even plate the dish beautifully.

The paper you shared is essentially a Chef's Safety Manual for this new kitchen. It asks: "If we let students use this magical robot, how do we make sure they still learn to cook, and how do we stop them from just handing in the robot's work as their own?"

Here is the paper broken down into simple, everyday concepts:

1. The Problem: The "Magic Wand" vs. The "Muscle"

The authors explain that AI is amazing. It saves time and helps you understand difficult concepts (like a GPS helping you drive). But, if you let the GPS drive the car for you, you never learn how to steer.

The Risk: If students just ask the AI to write their code or essays and submit it, they aren't building the "muscle" of critical thinking. They might get an 'A', but they won't know how to fix a broken engine later in life.
The Goal: We don't want to ban the GPS. We want to teach students how to use it as a tool, not a crutch.

2. The Solution: New Rules for the Kitchen

The paper suggests that instead of banning the robot, teachers need to change how they test the students. They break this down into two parts: General Rules and Specific Rules.

The General Rules (The "Golden Rules" for all classes)

The "Receipt" Rule (Disclosure): If you use the robot, you must write down exactly what you asked it to do. It's like showing your receipt at a store; you can't hide that you bought the ingredients.
The "Taste Test" (Oral Defense): After a student turns in a project, the teacher asks them questions about it. "Why did you choose this spice?" "How does this code work?" If the student can't explain it, they didn't cook it.
The "Remix" Assignment: Instead of asking for a final dish, ask the student to let the robot make a draft, and then have the student fix it, improve it, and explain why they changed it. The grade is for the improvement, not the original draft.

The Specific Rules (Tailored to the Task)

The paper gives specific advice for different types of "dishes":

Homework: Don't ask vague questions like "Explain sorting." Ask specific, weird questions like "Sort this specific list of 1,000 numbers using this specific rule." The robot might get it wrong, forcing the student to think.
Coding Projects: Require students to keep a "cooking diary" (version control logs like Git) showing every step they took. If they can't show the history of how they built it, they didn't build it.
Essays: Have the student write an essay with the AI, and then write a second essay without it. Then, ask them to compare the two. The grade is for their ability to spot the differences and fix the AI's mistakes.
Oral Presentations: You can use AI to write your script, but you have to answer the audience's questions live. If you freeze when asked a question, the script didn't help you learn.

3. The "Math Recipe" (The Formal Model)

The authors get a bit technical in Section 5, but think of it as a Scorecard Formula.

Imagine a student's final grade isn't just one number. It's a smoothie made of three ingredients:

The Final Product (40%): How good does the code/essay look?
The "Remix" Effort (30%): How much did the student improve the AI's draft?
The "Taste Test" (30%): Did the student pass the oral interview proving they understand it?

There is also a Safety Switch (Integrity Flag). If the student didn't admit they used the robot, OR if they failed the oral interview, the safety switch flips to "Zero," and they get no credit, no matter how good the final product looks.

4. The Big Picture

The paper concludes that we can't stop the AI train; it's already moving. Instead of trying to stop it, we need to build better tracks.

Old Way: Ban the robot, hope students don't use it, and punish them if they do.
New Way: Embrace the robot, but change the tests so that using the robot requires the student to think harder, explain more, and prove they are the master chef, not just the robot's assistant.

In short: The paper is a guide for teachers to stop fighting the future and start teaching students how to use the future responsibly, ensuring that when they graduate, they are the ones holding the steering wheel, not the AI.

1. Problem Statement

The rapid integration of Generative AI tools (e.g., ChatGPT, GitHub Copilot) into computing education has created a dual impact: while these tools enhance learning efficiency and self-efficacy, they pose significant threats to academic integrity.

Core Issue: The absence of structured guidelines leads to the misuse of AI, where students submit AI-generated work as their own, resulting in superficial understanding, a lack of critical thinking, and potential plagiarism.
Context: Traditional assessment methods (homework, take-home exams, coding projects) are increasingly vulnerable to AI automation. Institutions face a dilemma between banning tools (which ignores their pedagogical potential) and allowing unrestricted use (which compromises educational outcomes).
Specific Challenge: There is a lack of a unified framework that balances the benefits of AI assistance with the imperative to verify student authorship and conceptual mastery in computing disciplines.

2. Methodology

The paper employs a qualitative analysis and framework design approach, structured as follows:

Classification of Assessments: The authors categorize common assessment types in computing colleges (e.g., homework, take-home exams, in-class quizzes, coding projects, design projects, oral presentations, peer assessments) to analyze how specific AI tools (LLMs vs. Code-completion tools) impact each.
Guideline Development: Based on the classification, the authors derive a two-tiered set of guidelines:
1. General Guidelines: Applicable across all assessment types (e.g., mandatory disclosure, oral defenses).
2. Specific Guidelines: Tailored strategies for each assessment type (e.g., requiring version control logs for coding, comparative analysis for essays).
Formal Mathematical Modeling: The authors propose a formal mathematical model to quantify the evaluation process. This model defines sets for assessment tasks, tools, and guidelines, and introduces functions for disclosure, refinement, verification, and grading.

3. Key Contributions

A. Taxonomy of AI Impact on Assessments

The paper provides a detailed mapping of AI tool capabilities to assessment risks (Table 1):

Text Generators (ChatGPT): Primarily affect essays, literature reviews, and conceptual homework. Risks include superficial understanding and lack of originality.
Code Assistants (GitHub Copilot): Primarily affect programming assignments and projects. Risks include code outsourcing and dependency without comprehension.

B. Comprehensive Guideline Framework

The paper offers actionable strategies for instructors:

General Guidelines:
- Mandatory Disclosure: Students must explicitly cite AI usage and explain the extent of assistance.
- Verification Mechanisms: Implementation of post-submission interviews, oral defenses, and live coding sessions to verify authorship.
- Critical Review: Assignments where students must critique, modify, or enhance AI-generated drafts.
Assessment-Specific Strategies:
- Homework: Use specific, parameterized problems (e.g., unique datasets) to prevent generic AI solutions.
- Programming Projects: Require version control logs (Git), design rationales, and iterative code walkthroughs.
- Literature Reviews/Essays: Require a "Comparative Evaluation Report" where students analyze the differences between an AI draft and their own independent writing.
- Oral Presentations: Focus on Q&A sessions to test spontaneous understanding rather than script reading.

C. Formal Mathematical Model for Evaluation

The paper introduces a rigorous mathematical framework to evaluate AI-assisted submissions.

Definitions:
- $\mathcal{A}$ : Set of assessment tasks.
- $\mathcal{T}$ : Set of approved AI tools.
- $S_{AI}$ : AI-generated draft; $S_{ST}$ : Student-authored submission.
- $d(a)$ : Disclosure flag (0 or 1).
- $r(S_{AI}, S_{ST})$ : Refinement factor (0 to 1), measuring student modification of AI output.
- $v(S_{ST}, r)$ : Verification flag (0 or 1), confirming understanding via oral/interactive checks.
Grading Function:
$G(a) = w_1 \times \text{score}(S_{ST}) + w_2 \times r(S_{AI}, S_{ST}) + w_3 \times v(S_{ST}, r)$
Where $w_1, w_2, w_3$ are weights summing to 1, adjustable based on the assessment type.
Integrity Flag:
$I(a) = d(a) \land v(S_{ST}, r)$
The assessment is only valid ( $I(a)=1$ ) if the AI usage was disclosed AND the student passed the verification of understanding.

4. Results and Findings

Feasibility of Integration: The study demonstrates that AI tools can be integrated into computing education without compromising integrity, provided that assessment designs shift from "output-only" evaluation to "process-and-understanding" evaluation.
Effectiveness of Verification: The proposed model highlights that oral verification and refinement analysis are the most effective mechanisms to detect misuse and ensure learning.
Equity Considerations: The paper notes that guidelines must address equity, ensuring assessments do not rely on premium AI features unavailable to all students, and that human oversight is required to mitigate algorithmic bias.
Workflow Optimization: Figure 2 illustrates a streamlined workflow where AI usage is transparent, submissions include analysis/reflection, and final grading incorporates verification outcomes.

5. Significance

Practical Framework for Educators: The paper moves beyond theoretical debate to provide a concrete, operational toolkit for instructors in computing colleges. It offers specific rubrics and strategies for diverse assessment types.
Standardization of Integrity: By introducing the Integrity Flag ( $I(a)$ ) and the Refinement Factor ( $r$ ), the paper proposes a standardized way to mathematically define and calculate academic honesty in the age of AI.
Pedagogical Shift: It advocates for a paradigm shift where AI is treated as a "calculator" or "scaffold" rather than a replacement for thought. The focus moves from preventing AI use to teaching responsible AI engagement and critical evaluation of AI outputs.
Scalability: The formal model allows institutions to adapt grading weights ( $w_1, w_2, w_3$ ) based on specific learning outcomes, making the framework flexible for various computing disciplines (e.g., software engineering vs. data science).

In conclusion, the paper argues that the solution to AI-induced academic dishonesty is not prohibition, but structured integration guided by transparency, critical engagement, and rigorous verification mechanisms.