AI Use Policy: Critical Thinking and Research Quality

Last update: 2025-12-03

Table of Contents

Purpose and Core Principle

Generative AI (GenAI) and Large Language Models (LLMs) are powerful tools that can accelerate research, but only when used ethically, critically, and responsibly. This policy defines acceptable and unacceptable uses of AI to ensure research quality, academic integrity, and the professional development of all lab members.

Core Principle: GenAI is a highly capable assistant, but it is not a substitute for your expertise and should never hinder your development as a researcher.

If AI could do your expected research tasks faster and cheaper than you, then your role would be obsolete. The core value of a research student lies in critical thinking, problem-solving, and developing novel approaches. Your value is rooted in what the AI cannot do: the deep analysis, the ethical judgment, the validation, and the persistence to solve truly difficult problems.

General Principles of Accountability

Ultimate Responsibility: You are solely and fully responsible for all final outputs, whether AI-generated or not. This includes the correctness of code, the validity of scientific results, and the accuracy of all written material.
Critical Evaluation is Mandatory: Never accept AI-generated content (text, code, data analysis, etc.) without a thorough review. You must verify correctness, coherence, and validity using your domain knowledge.
Prerequisite Knowledge: AI is most effective when used by an expert. You must possess sufficient fundamental knowledge and experience in the subject matter to effectively supervise, guide, evaluate, and validate the AI’s output. Do not blindly trust GenAI.

Specific Rules for Scientific Writing (Papers, Reports, Proposals)

Action	Status	Mandatory Conditions & Verification
Generating scientific text from scratch (e.g., full paragraphs, results, discussion)	PROHIBITED	N/A
Improving clarity of text (grammar, spell check, style refinement of your own writing)	ALLOWED	You must critically evaluate all suggested edits to ensure the scientific meaning or intended tone has not been altered.
Initial Tasks (Brainstorming, summarizing literature, outlining)	ALLOWED	All information, claims, and references must be independently verified as real and accurate.
Citations and References	LIMITED	Verify all generated references. GenAI frequently “hallucinates” non-existent citations. You must physically check every source, not only that it exists, but also that it contains the information the AI claims.

Specific Rules for Coding & Scientific Implementation

Boilerplate & Routine Code, Basic coding tasks:
- You may use GenAI to generate simple, repetitive, or syntactic code (e.g., file parsers, basic utility functions, plotting, simple API calls).
- You may use AI for basic and routine coding tasks, such as getting hints or explanations for debugging, automating routine tasks, and refactoring code.
- Condition: You must understand every line of code used in your research. You must validate correctness of code generated by AI.
Complex Algorithms:
- You must not rely on GenAI to write complex algorithms, scientific computing methods, or novel research-specific implementation logic from scratch. You must write this code yourself.
- You may use AI’s assistance in implementing small steps of complex code, but you must carefully and rigorously supervise and validate its generated code. This requires expertise and experience in both the subject matter and programming, usually acquired only through writing code yourself without AI’s assistance.
Rigorous Validation: If you use AI-generated code, you must execute rigorous testing and critical analysis. If the code runs but produces incorrect scientific results, it is a failure of your validation process, not the AI’s.
No Giving Up: If GenAI fails to implement a complex method correctly, you are responsible for learning and implementing the solution yourself. AI limitations are not an excuse for incomplete work or abandoning coding tasks.

Consequences and Enforcement

Maintaining the lab’s scientific integrity is non-negotiable. Failure to check AI output is considered negligence. If incorrect work occurs because you blindly trusted AI, failed to check results, or relied excessively on AI for prohibited tasks:

Loss of Privilege: You may lose the privilege to use AI tools for coding and/or writing in your research until you can demonstrate responsible use.
Mandatory Review: Your work may be subjected to mandatory, increased scrutiny (e.g., additional code review or writing revision requirements).
Continuation Discussion: Repeated issues will result in a formal discussion regarding your continuation in the research program, as it reflects a fundamental lack of commitment to essential research skill development.

Summary

Use AI as a helper, not a crutch.

Think first, verify always, and never outsource your expertise.

Your job is to be the researcher-AI is just a tool.