In the Age of AI, What Does a University Thesis Really Prove?

A polished thesis is not the same as demonstrated understanding. That distinction will frame a student debate at UJK in Kielce, Poland, on November 18, 2026. The university’s first Artificial Intelligence Day will conclude with an Oxford-style debate on whether the traditional degree thesis has lost its purpose in the age of AI. This is a debate proposition, not an announcement that UJK is abolishing theses.
The question deserves attention beyond one campus. Rather than choosing between unrestricted automation and a return to examination halls, universities should ask a more useful question: What evidence should a degree require when students can produce substantial work with AI assistance?
For university leaders and education technology teams, the opportunity is to redesign assessment around research decisions, demonstrated understanding, and accountable tool use. The thesis can remain central, but its final manuscript should not carry the entire burden of proving competence.
Preserve the Research, Rethink the Evidence
The strongest case for changing the thesis is not that writing has become worthless. It is that a completed document may provide insufficient evidence of what its author understands. Assessment redesign therefore needs to connect the final product with oral explanations, practical demonstrations, reflections, and successive revisions. These are central elements of RRC Polytech’s approach to assessment in an AI-enabled environment.
A useful redesign should begin with learning outcomes, not a list of prohibited tools. Before deciding where AI belongs, a department should specify what students must demonstrate:
Can they formulate a defensible research question?
Can they select and justify an appropriate method?
Can they distinguish reliable evidence from unsupported claims?
Can they explain limitations and respond to counterarguments?
Can they take responsibility for the final conclusions?
These questions suggest a different assessment structure: retain substantial research work, but require evidence at selected points throughout its development.
This does not mean documenting every keystroke. A proportionate model could collect a research proposal, a methodological decision record, selected revisions, and a final defense. Each component should serve a defined assessment purpose rather than create administrative work.
The goal should be visible reasoning, not continuous surveillance.
Combine Open Research With Secure Demonstration
A practical framework separates opportunities to develop knowledge from opportunities to verify it.
The University of Sydney’s two-lane approach makes this distinction explicit. Open assessments support learning with contemporary tools, while secure assessments establish whether students have achieved the required outcomes. Secure does not automatically mean AI-free: controlled AI use can be appropriate when responsible tool use is itself being assessed. The framework also allows an open written submission to be paired with a secure oral assessment.
Applied to a thesis, this suggests three connected components:
The Research Submission
Students produce a substantial written argument, supported by evidence and discipline-appropriate methods. The assignment specifies where AI assistance is permitted, restricted, or irrelevant.
The Process Record
Students submit selected evidence explaining how the work developed. This might include source evaluations, experimental records, analysis code, design iterations, or explanations of significant changes.
The Individual Defense
Students explain their choices and respond to questions grounded in their own submission. Examiners might ask why a method was appropriate, how a conclusion follows from the evidence, or what would change under different assumptions.
These components should assess complementary capabilities. A strong manuscript should not automatically compensate for an inability to explain its central argument.
Equally, an oral defense should assess understanding rather than presentation confidence. Departments should plan accessible arrangements, consistent questioning principles, and clear marking criteria from the outset.
Assess Contribution, Not Just Text Production
Thesis redesign also needs a more precise account of authorship.
A mixed-methods study covering 2,441 students with prior generative-AI experience at 33 Chinese art and design institutions illustrates the distinction. In follow-up interviews, 21 of 22 participants distinguished control over conception and judgment from delegating execution. However, the convenience sample excluded AI non-users and was geographically concentrated, so the findings should not be generalized to all students or disciplines.
For assessment design, the useful implication is to ask students to explain their intellectual contribution, rather than disclose tool use through a generic checkbox.
A proposed contribution statement could address:
Conception: Who developed the research question and core argument?
Delegation: Which tasks involved AI assistance?
Verification: How were outputs checked against evidence?
Judgment: Which suggestions were rejected, revised, or retained, and why?
Responsibility: What can the student independently explain and defend?
The evidence should vary by discipline. A software thesis could require the student to defend architecture choices, demonstrate tests, and explain security trade-offs. A history thesis could emphasize source provenance and competing interpretations. A design thesis could examine how concepts evolved through selection and revision.
These are proposed assessment patterns, not evidence that one format will work everywhere. Departments should test them against their own learning outcomes.
Build Institutional Capacity Before Adding Technology
Assessment reform requires staff time and support. RRC Polytech’s five-hour blended AI Assessment (Re)Design Lab offers one operational example: approximately 133 instructors participated across its first eight deliveries in spring and summer 2026. That figure demonstrates participation, not independently measured improvements in student learning.
Universities considering similar changes should combine faculty development with a restrained technology strategy.
For implementation teams, a sensible starting architecture would connect assessment requirements to existing learning systems:
Structured milestones: Capture proposals, selected drafts, and supervisor feedback.
Contribution records: Provide concise, assignment-specific disclosure fields.
Controlled access: Limit sensitive research materials to authorized users.
Portable evidence: Allow students and departments to export assessment records.
Clear retention rules: Define what is stored, why it is needed, and when it is deleted.
Institutions should avoid requiring complete AI conversation histories by default. Instead, they should identify the minimum evidence needed to assess each outcome and provide a safe route for handling confidential or protected material.
A pilot should evaluate more than submission quality. Useful measures could include students’ ability to defend decisions, marking consistency, accessibility, staff workload, and clarity of AI-use expectations.
The November 18 debate offers a timely starting point, but the strategic question is not whether universities should abandon the thesis. It is whether they can make the thesis a stronger demonstration of capability.
The most credible direction is to preserve demanding research while making understanding, judgment, and responsibility explicit. A degree should certify what a graduate can explain and defend, not simply what they can submit.



