AI can help a doctoral student develop and test a research idea much faster than before. I welcome the scope this gives young researchers to attempt more ambitious work. Yet the route to a result has always been part of their education. Struggling with an unsuccessful approach can reveal a hidden assumption or suggest a question that the original project never considered. As more of this work is delegated, universities need to consider how such understanding will develop.
In their September declaration “A Severe Misalignment of AI in Mathematics”, a group of Fields Medallists argued that solving problems serves the deeper purpose of developing understanding and the capacity to formulate new ideas. Their concern is directly relevant to doctoral students, who are learning to shape and conduct research of their own.
Scientific judgement begins before a result exists. It guides the choice of a problem and remains essential when we decide what the evidence supports.
The European Commission’s “Living Guidelines on the Responsible Use of Generative AI in Research”, developed with contributions from CESAER and updated in October 2026, emphasise researchers’ responsibility and oversight of critical decisions. Doctoral education must develop the expertise needed to exercise that oversight.
Consider a student faced with an unexpected result. They might develop an explanation of their own, use AI to explore alternatives, and design and carry out a test to distinguish between them. A discussion with the supervisor can then examine how the outcome changes the interpretation, including whether an initially attractive explanation still holds. The student should feel able to explain which steps involved AI and where uncertainty remains. Working through such cases with colleagues can also help supervisors refine their practice.
If AI saves time, we should allow some of it to be spent on a difficult question whose answer is uncertain. Progress reviews should recognise what a student has learned from carefully conducted work, even when it has not yet led to a publication. Expectations that simply rise with the speed of the tools could leave doctoral students with less room to think.
Creating these conditions for doctoral education is central to Wrocław Tech’s ambition to join Europe’s leading research universities. Our transformation plan has been endorsed by international experts, and we have recently been selected as one of ten universities in Poland’s Excellence Initiative – Research University (IDUB) programme.
Our Doctoral School has now grown to around 1,000 doctoral students: roughly two for every five members of academic staff and one for every twenty other students. This is a substantial doctoral community by Polish standards. In relative terms, however, doctoral students still account for a smaller share of our student population than at many European research-intensive technical universities.
We intend to grow further while strengthening the quality of doctoral education. Growth should give students more opportunities to learn from one another’s experience of using AI. Supervisory capacity must also keep pace, so that supervisors have time to follow how each student’s reasoning develops.
Within the Doctoral School, mini-grants give students resources for their own projects and research visits. A code of good practice sets expectations for supervision, while reduced compulsory teaching in the final year protects research time.
Wrocław Tech, one of Poland’s largest universities, brings together science, engineering, medicine and the social sciences. A doctoral student developing an AI model with clinical data, for example, can draw on this expertise while defining the project. A readily available dataset can otherwise draw a project towards what is easiest to measure. Clinicians can help establish which outcomes matter for treatment. Social scientists can examine how practitioners interpret the model’s output and how working conditions affect its use in clinical decisions. These contributions help determine the model’s purpose, the data collected and how its reliability should be tested in other patient populations. In shaping the project, the student learns what evidence is needed to connect technical performance with a claim about clinical value.
Interdisciplinary learning can also take place beyond the project team. The Academia Europaea Wrocław Knowledge Hub, which we host, offers seminars where doctoral students can see how scholars in other fields build an argument and respond to criticism.
Alongside intellectual exchange, researchers need practical support for using AI in their own work. Our Faculty of Information and Communication Technology educates around 5,000 students, and the university operates its own supercomputing centre. We coordinate CLARIN-PL, Poland’s contribution to the European language research infrastructure, and led the original consortium that developed the PLLuM family of Polish language models.
Our transformation plan aims to make this expertise and infrastructure available to every researcher, including doctoral students, whatever their discipline. We envisage a shared AI, data and high-performance computing platform with practical training and support from research software engineers and data specialists. Researchers would learn to select suitable tools, develop reproducible workflows and protect confidential data through work on their own projects. Doctoral students should participate in designing and testing these workflows, so that specialist support helps them use AI effectively and safely while becoming more independent in their research.
Through the Unite! European Universities alliance, our doctoral students can work with supervisors and research groups at partner institutions. This experience exposes them to different research traditions and approaches to supervision. As a new Member of CESAER, we would welcome extending further the peer-learning activities of its Task Force European Higher Education to doctoral research. A shared collection of AI-assisted research cases, showing which decisions doctoral students made themselves and how they justified them, could complement CESAER’s recent work on AI skills and competences while helping to improve doctoral supervision and assessment.
When a researcher completes a doctorate, we should be able to ask: what will you investigate next, and why? AI can help make the next question more ambitious. It remains our job to educate someone who can pursue it with sound judgement.
By Arkadiusz Wójs, Rector of Wrocław University of Science and Technology
We use both our own cookies and those of carefully selected partners we collaborate with.
Check out our detailed Cookie policy » and our Privacy policy » .