Responsible use of generative AI in doctoral research

When integrating Generative AI into your PhD research, remember that you, as a PhD researcher, are not operating in a vacuum. One of the most crucial best practices is to proactively discuss your AI strategy with everyone involved in your PhD project, as perspectives and rules can vary significantly.

It is highly recommended to establish clear agreements on (Gen)AI usage early in your trajectory with:

• Your Supervisor(s): Ensure you share a common vision with your supervisor about the use of gen AI during your PhD project. Ask your supervisor about their view on the use of AI. Agree exactly on if and for which purposes AI may be used (e.g., text editing, content generation, literature review, methodology planning, coding,...). If critical questions about your AI use should arise during a doctoral committee meeting or your final PhD defense, having your supervisor’s full backing is essential to justify your choices.
• Doctoral Committee & Jury: Be transparent during your annual evaluations. Knowing their stance helps you appropriately justify your methodological choices and build trust.
• Funders: Granting agencies (such as FWO, BOF, or European funding agencies) often have strict, evolving policies regarding data privacy, intellectual property, and AI-generated content. Non-compliance could jeopardize your grant.

Open communication prevents misunderstandings down the line. By openly discussing your AI tools and methods, you protect your scientific integrity and ensure your research approach is fully supported by your entire academic network.

When using (Generative) AI tools, safeguarding your research data must be a top priority. Entering data into an AI prompt can be equivalent to making it public, depending on the tool’s privacy policies.

To protect your intellectual property (IP), ensure the privacy of your research subjects, and comply with GDPR, follow these essential guidelines:

• Never upload sensitive data to public or free AI tools: Standard versions of tools (like the free version of ChatGPT without opt-out) often use user input to train their models. Do not input personal data, confidential records, proprietary code, or unpublished research findings into these platforms.
• Use secure, institutionally approved tools: UHasselt strongly recommends using secure AI environments where data protection agreements are in place. Always check the UHasselt AI-tools webpage for the most up-to-date, secure options.
• Anonymize or pseudonymize: If you intend to use AI for text editing or structuring, thoroughly strip all identifying information, critical variables, and confidential details from your prompts beforehand.

Always review the privacy settings of any AI tool you use and actively opt out of data training whenever possible

While Generative AI can accelerate your workflow, it lacks true comprehension. It can confidently present completely false information ("hallucinations") or subtly reproduce systemic biases present in its training data. As a PhD researcher, rigorous verification is your primary defense against compromising your scientific integrity.

Follow these verification practices:

• Always cross-reference: Never accept AI-generated facts, citations, or data summaries at face value. Manually verify every claim, statistic, and reference against primary, peer-reviewed literature.
• Watch for hidden biases: AI models often skew towards specific demographics, English-centric perspectives, or overrepresented methodologies. Actively evaluate whether the output overlooks diverse viewpoints or introduces unverified assumptions into your research design.
• Test the logic, not just the language: AI is excellent at sounding highly plausible. Deconstruct AI-generated arguments, structures, or code step-by-step to ensure they are scientifically correct and logically coherent.

Remember, you are the domain expert and the final responsibility for the accuracy and objectivity of your doctoral thesis rests entirely with you.

The world of (Generative) AI is moving very fast. Tools that are cutting-edge today might be outdated in a few months, and new ethical and methodological standards are constantly emerging. As a PhD researcher, it is your responsibility to stay informed and continuously educate yourself.

Make active learning a core part of your professional development strategy:

• Leverage Doctoral School offerings: Take advantage of specific AI workshops organized by the UHasselt Doctoral Schools (such as sessions on using GenAI for research, processing sensitive data, or AI in science communication) as well as training offered by partner networks like FLAMES or VAIA.
• Follow policy updates: Regularly check the UHasselt university-wide landing page for (Gen)AI. Institutional policies, recommended tools, and data security guidelines will evolve, and the most up-to-date information will be available on this webpage.
• Engage with your community: Discuss new tools and prompt engineering strategies with your peers and within your research group. Sharing experiences and pitfalls helps the entire academic community stay critical and agile.

With the increasing integration of (Gen)AI in the writing process, the traditional approach to evaluating a doctoral thesis is evolving. Because AI can significantly polish language and structure written content, academic language and style should remain a requirement, but weigh less in the evaluation.

An important good practice is to shift more evaluative weight toward the oral examinations — specifically, the doctoral committee's yearly meeting, the internal pre-defense, and the public defense. For faculties/schools that do not yet standardly organize an internal pre-defense, introducing this closed-door examination is highly recommended as a best practice. Along with the final public defense, these interactive assessments serve as a good test of true knowledge and ownership.

During the committee meetings, the pre-defense and public defense, members are encouraged to actively probe beyond the written text. By asking unscripted, in-depth questions about methodological choices, data interpretation, and the specific ways AI was utilized during the research cycle, the members can accurately assess the PhD researcher’s critical thinking and domain expertise. This oral dialogue ensures that the PhD researcher fully understands the core science, can defend their findings in real-time, and proves they are the true architect of the work, rather than just a skilled AI prompt engineer.

The landscape of AI in academia has shifted significantly over the last couple of years. Major funding bodies have moved from vague guidelines to concrete policies to protect scientific originality, data security, and the integrity of the peer-review process. An overview of good practices is given below:

• Align with the guidelines of the funding agency regarding AI
• Protect your IP and data
• Retain your authentic voice - do not use AI to write your grant
• Fact-check everything

Interesting link: https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1011863#sec002