The Experience of Erasmus+ Music Students Fishing in the Murky Pond of GenAI

Angela Siobhan Wright

¹ Head, Teaching & Learning Unit, AnSEO – The Student Engagement Office, Arts Office, Office of VP for Academic Affairs & Registrar, MTU, Ireland

How to cite: Wright, A. S. (2026). The Experience of Erasmus+ Music Students Fishing in the Murky Pond of GenAI, In: 12th International Conference on Higher Education Advances (HEAd’26). Valencia, 15-18 June 2026. https://doi.org/10.4995/HEAd26.2026.21754

Abstract

Building on previous research on post-graduate supervision at both masters and doctoral level and the experiences of post-graduate students with implications for Higher Education, this work presents a unique perspective of an Erasmus+ group on a Blended Intensive Programme to the researcher’s university. The participants are high calibre musicians of international standing pursuing a masters by research programme in their European university. The aim of this research is to explore their unique perspectives on using a GenAI tool for their research thesis. Results are insightful, with varying outcomes to the exploration of utilising an open GenAI tool. The main results opined by the participating students are, lack of trust, lack of depth with an overall conclusion that use should be approached with caution and human oversight. The main conclusion is that students are very savvy and understand the risks of such tools when it comes to their own research.

Keywords: Generative Artificial Intelligence, Blended Intensive Programme (BIP), Post-Graduate Research, Research Supervision

1. Introduction

This research paper builds on previous empirical work (Wright, 2020;2024;2025a;2025b) on ‘best practice’ for post-graduate research supervisors in the context of new challenges faced by Artificial Intelligence and Generative Artificial Intelligence (GenAI) for Higher Education. Prior work explored the challenges for supervisors and students in one university setting, along with the support staff of administration, library and management staff with over 170 participants in all. This new research explores the perceptions for a masters cohort of adult learners specifically in the area of the creative arts who visited the researcher’s university from other European universities. This group was involved in an Erasmus Blended Intensive Programme (BIP) project and visited the host country for a week. This collaboration involves an Irish host university, with European partners from Kunst Universität Graz (Austria), LUCA School of Arts Leuven (Belgium) and the Royal Conservatoire (Netherlands). A BIP is acknowledged as a forward-looking approach to higher education (Crespo & Prodanović, 2025). Erasmus+ BIPs were introduced by the European Commission in 2021, providing innovative approaches to higher education mobility (European commission, 2022; Frank, Schweighofer, 2026). This was achieved with the combination of physical mobility of between five and thirty days, together with mandatory online collaborations that can take place at any stage of the BIP process. An award of a minimum of 3 ECTS credits is given to participants. A requirement of the programme is that students or staff from at least three HEIs across at least three different Erasmus+ countries participate. The physical component normally occurs at the host institution, or another location in its country. Participants receive Erasmus+ funding through their home institutions. Host institutions obtain support for coordinating the physical element (European Commission, 2022; Frank, Schweighofer, 2026).

The novelty of this research is in the unique cohort of musical performers and researchers that travelled to the host country and participated in the BIP. This particular BIP programme explored the connections between practice-based research, teaching and performance. All participants were performers of a musical instrument as well as academic researchers. The strength in this gathering and connections across European universities is examining best practice and aligning approaches for masters’ students from the visiting institutions. These sessions enabled dialogue and opportunities to co-create and learn from each other while exploring concepts and applying these to personal research practice and practical performance. These concepts of shared experiences were key to this programme and visit, so that participants would return home with novel ideas inspired by discussions, investigation and exploration that would enhance and develop their future research skills. As part of this programme, a workshop on best practice for postgraduate research was undertaken with this researcher to explore approaches to, and views on, Artificial Intelligence tools in research.

2. Methods

The workshop explored the understanding and interpretation of working with Generative Artificial Intelligence in a music research context. It is important to note that this test, carried out in the context of the workshop, was to experiment with one generic AI tool for all in the room using an OpenAI tool (Open AI GPT-5.1). The researcher acknowledges that other AI tools can be considered more specific in a research context; however, the aim here was to observe what the students found when they all searched at the same time, with the same prompts for their themes. For this, a qualitative approach was applied to interpret the views of Twenty-two masters’ students (15 Female, 7 Male) who contributed to the study and workshop. The approach was observational together with in-depth discussions after a number of tests were undertaken using the OpenAI tool. Ethics approval was granted for the research prior to the workshop, and the participants were fully informed as to their participation and contributions and their opt-out options and rights and the integrity of the work. All participants agreed to the research as they saw it as an interesting experiment. All participants remained anonymous and are only counted in terms of their groups. First, the researcher presented a transparent overview of the challenges of GenAI in the research context for researchers and for research students. Then the students were divided into 4 groups (2 groups x 4 + 2 groups x7) where they were asked to work together on the following task:

Using the OpenAI tool [https://openai.com/] GPT-5.1 generation in your groups ask the tool to perform the following task:

1. Write a 6oo word literature review on the difference between the experience of being at a live musical performance and viewing a recorded broadcast performance [Group1]

2. Write a 6oo word literature review on human music composition and Artificial Intelligence music composition [Group 2]

3. Write a 6oo word literature review on the future of music performers in the context of Artificial Intelligence [Group 3]

4. Write a 6oo word literature review on the difference between using real instruments on recordings and AI generated Musical/Vocal instruments [Group 4]

Each student in each group was asked to run the same experiment on their individual computers, for example, if there were 4 participants in a group, they ran the same experiment 4 times; if 7, they ran it 7 times. This proved insightful.

After the first question from the OpenAI tool, all participants in the groups were asked to reflect on their experience using the following 4 themes and writing their impressions on their worksheets under the headings of these themes below:

1. Outline their impressions of the results.

2. Relevancy of the output

3. Accuracy of the output

4. Reflections/ Actions/ Next Steps to Prepare

Then they were asked to repeat the question again with the same OpenAI tool to investigate the consistency of the results and to check and compare the outputs for similarities, differences etc. The researcher asked participants to use the free tool to ensure equality and explore what research students who may not have access to paid models of AI would experience. Each time, participants were asked to document their reflections on the worksheets using the 4 same themes. Intensive discussions followed, with participants then tasked with asking alternative questions of the AI tool, like “please add extra citations” from specific years, focusing on specific countries, specific disciplines etc., to explore the outputs.

At the end of the workshop, the groups summarised their impressions and views of the experiment on handwritten worksheets where open expression was encouraged under the 4 themes above. The data was gathered from the immersive experience of being in the room with the students and observing their reactions to the outputs, and then, afterwards, the information on the students’ impressions from the worksheets was analysed by the researcher by hand, reading and re-reading the comments from the workshop and collating all comments. The data was extracted using the themes and the comments grouped into the relevant areas using the (Braun & Clarke, 2006) approach of thematic analysis. Analysis began with the first pass approach to get an initial feel for the outputs, then, the worksheets were carefully read to identify similar threads and themes to find meaning. These were then coded and grouped into the important pieces using colour coding (full coding is detailed outside of the confines of this paper). Concrete outputs emerged within the data.

3. Main Findings & Observations

The first interesting element of this research was the fact that not all participants had used an OpenAI tool; 2 participants of the 22 had not. Their reactions were, “this is scary”; “unbelievable”, “almost magical”. Participants that had used a tool previously, also used the words, “this is really scary”, “it is so fast”, “a strange experience”. When it came to discussing the initial outputs, participants noted the format was very specific, and it mostly consisted of citations with a lack of depth in terms of the body of the output (7 participants). For some, all the citations existed, for others, they were inaccurate; for example, 13 citations in the reference section with only 10 cited in the body for one group. One student described the citations as ‘shady’ and wondered how relevant and important the output was. One group when trialing and re-trialing received different citations for the exact same topic in the 4 different versions. Other observations from this group were, “similar layout with slightly different words, “very generic text” and not “really to the point”. Other criticisms of the citations were that some “were very old”, for example, “1963, 1957” (3 groups). One group observed that when they asked for additional citations the tool just added more citations to the reference list and did not embed the information in the body of the main work. Sources were “detached from and not joined up in the topic”. “The output seemed repetitive and could have been synthesized much better”.

For other checks, all citations were correct, so there was no consistency. In another group, the citation output was “impressive and very relevant with key citations”. The format of the outputs was discussed with the participants noting “this is nicely laid out” (4 participants), nicely structured and without spelling mistakes (7 participants). As English was not a first language for participants, discussions ensued around this benefit. Negative feedback for example, in one group with 7 participants, consisted of outputs that were very similar which led to the discussion around whether, if this work was submitted as an assignment, would plagiarism be an issue along with accusations from supervisors or lecturers around similarity. Another group outlined that when they tested and retested, the output seemed to discuss the same topics and the same sources but with a different tone and voice. Other observations were for some outputs, “titles were in miniature” making it difficult to read and off putting. Another group output contained a conclusion which they described as “weird”. Another group criticized their return as “it lacked a conclusion”. Another group strongly discussed that the output was very evidently non-human generated and that they would “notice this if they received this submission to grade”, with another student outlining that their AI outputs “were not human like”. Another group were very clear that you “must check the citation outputs with human oversight”. They found the citations to be inaccurate and wrong, for example, “incorrect years, names”. Another group found that when they checked the word count output it was incorrect, which led to much humour and discussion on the fact that “it couldn’t even count” (word count 746 for 600-word request); another group checked theirs and found it was under the word count. Another student said, “I hate it”, I really do”, “the words are without semantic depth”.

In terms of overall learnings from the workshop, students outlined that they would now be more critical of what they read, and careful of citations and articles they sourced online. Comments such as “I don’t like it because it is not me, or my own work”. “It can be useful for making connections, but it lacks critical thinking and depth”, “I feel the outputs are random and not what I would like to contribute as a researcher”, “I don’t like the random data scraping”, “For me after today, I would prefer to avoid it”. “It lacks the critical links and connections that we do when we research deeply”. Another group summary was that it “delivered nothing that was groundbreaking, only lots of words with no real meaning or depth”.

4. Discussion

The findings from this unique study are interesting, building on the earlier work of (Dai et al., 2023; Li et al., 2023). There is no doubt that post-graduate supervision is changing, presenting some opportunities (Jensen et al., 2025) in terms of chatbots and feedback opportunities to students, but with some critical challenges due to the robustness and accuracy required by post-graduate outputs. Research supervision meetings may become easier with supervisors not having to face the challenging feedback sessions with their students (Wright, 2020); if we use bots, however, new challenges make the process testing. One thing is clear, human oversight is a vital component, and the substantial role of the supervisor can never be replaced. Universities are still in flux in terms of what is acceptable for post-graduate work while not morphing into the area of plagiarism. Trust and robust outputs are key challenges, with (Dai et al., 2023; Smith, 2026) noting that the narrative has changed in terms of ethics, trust, roles, and reliable outputs. Reflecting on the findings from this research cohort of 22 students, and reading Dai et al., (2023) who allowed their post-graduate students to have free reign in terms of AI tools, (that experiment ended shortly after their students became disillusioned with the outputs of these tools and did not trust them as reliable outputs for their important research work), gives this researcher hope for our young researchers that they are wary of the challenges and risks that AI tools pose. Some universities support the use of AI tools with complete disclosure; however, many academics do not agree with this approach leaving a ‘divide and conquer’ scenario among peers and post-graduate students. Some academics still hold that the use of these tools is in fact still plagiarism and believe that it goes against their own research values. So, where then can equality for all fit in this debacle? Now universities and faculties must find productive pathways through fully informed literacy for post-graduate students and their research supervisors. Responsible use of AI tools is critical. Arguments are made in the literature that there is a place for AI; however, students must be adequately supported with full literacy (Awdry, 2023; Smith, 2026). In an Irish context, the Higher Education Authority (HEA, 2025) published their recommended framework for generative AI use in higher education with clear pathways and responsibilities outlined, such as “human accountability cannot be delegated” and that university staff must be supported with “mandatory ongoing training” (HEA, 2025). Universities need clear “live mapping not accountability that is fuzzy” (Smith, 2026). Reflecting on the findings from the cohort of postgraduate students in this study their views relate directly to the HEA framework where decisions on outputs must and should be based on clear evidence and not assumptions and that AI adoption represents “managed compromise” (HEA, 2025). Smith (2026) supports this viewpoint advocating for the student voice in the process and that they must be taken seriously.

5. Conclusions

Attributed to the ancient Greek philosopher Heraclitus of Ephesus is the tenet: “the only constant in life is change”. Then, how we manage change and adapt best practice is a key reaction to this reality. However, the ancient approach to student supervision that we still apply today as a way of ensuring excellent and informed graduates with robust research impact is utterly demanding, wide-ranging (Taylor, 2019) and more complex than ever with new unimaginable challenges. Key challenges of inclusivity, pedagogical robustness, empowerment and environmental impacts are key challenges. From the findings of the students in this study, hallucinations, errors, biases and an inability to count are still evident. What now is our purpose as research supervisors as we fashion future generations of informed graduates to ensure an ethical and fair approach for all.

Critical thinking is key (Wright, 2020;2024,2025b) but will future students lose the ability to do this if they exclusively use an AI tool? Overall, one group summed up their experience, that the output lacked soul, felt empty, devoid of experience and was clearly scraped. It “feels fake, like I am stealing” and “doesn’t sit well with my values”, maybe “I will change my mind, but for now, I prefer the tried and tested approaches of read, review, summarize, move on”. Their collective viewpoint was that the whole experience jarred and ran counter to the artist’s hunger for authenticity, the need for the personal artistic journey, the individual prize of owning the intellectual and spiritual product, the creator’s magic and pride.

For now, what is the role for research, the research student, the research supervisor? Supervisors will have to rethink and rebalance and adapt. We need “cross-cutting oversight, not just scattered committees” (Smith, 2026). Conversations have shifted from ‘should we allow?’ to how can we manage this? How can we be fair? What is our purpose? How can educators and students be supported in an equitable manner? Maybe time will not be kind to us supervisors, or maybe time will be beyond all our hopes and aspirations. Shaping the future forces us now to stop, pause and ensure that, whatever the rubric, the work must adhere to ethical and legal parameters that will be fair to all who seek robust and purposeful research.

6. Disclaimer

No GenAI tools were used in the writing of this paper.

References

Awdry, R. (2023). A12 Making Friends with AI, N-TUTORR Academic Integrity Masterclass, March 23, Available at https://vimeo.com/815310897.

Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3(2), 77–101. https://doi.org/10.1191/1478088706qp063oa

Crespo, B., & Prodanović, M. (2025). Blended Intensive Programmes (BIPs) in Professional Development: An International Scenario for Training Staff. The Journal of Continuing Higher Education, 1–17. https://doi.org/10.1080/07377363.2025.2523608.

Dai, Y., Lai, S., Lim, C. P., & Liu, A. (2023). ChatGPT and its impact on research supervision: Insights from Australian postgraduate research students. Australasian Journal of Educational Technology, 39(4), 74–88. https://doi.org/10.14742/ajet.8843.

European Commission (2022). Blended mobility implementation guide for Erasmus+ higher education mobility KA131. Publications Office of the European Union (2022), Available at https://op.europa.eu/en/publication-detail/-/publication/8a4bbab0-540d-11ed-92ed-01aa75ed71a1/language-en.

Frank E, Schweighofer, P. H. (2026). Erasmus+ Blended Intensive Programmes' role in promoting inclusivity and sustainability in management education - A survey-based evaluation, The International Journal of Management Education, Volume 24, Issue 2, 101358, ISSN 1472-8117, https://doi.org/10.1016/j.ijme.2026.101358.

HEA. (2025). Generative AI in Higher Education in Teaching and Learning Policy Framework, HEA, Available at https://hea.ie/2025/09/17/generative-ai-in-higher-education-teaching-and-learning-sectoral-perspectives/.

Jensen, L. X., Bearman, M., Boud, D., & Konradsen, F. (2025). Feedback encounters in doctoral supervision: the role of generative AI chatbots. Assessment & Evaluation in Higher Education, 1–14. https://doi.org/10.1080/02602938.2025.2478155.

Li, M., Gibbons, J., Meng, H., & Taha, G. (2023). Graduate students’ experience on using ChatGPT in Education: A narrative inquiry. SSRN https://doi.org/10.2139/ssrn.4452108.

Smith, T. (2026). AI has moved into universities’ engine room, but no one is at the controls, Times Higher Education, Available at https://www.timeshighereducation.com/opinion/ai-has-moved-universities-engine-room-no-one-controls, January 19.

Taylor, S. (2019). A Framework for Good Supervisory Practice. UK: Council for Graduate Education. June, DOI: 10.13140/RG.2.2.12696.34564.

Wright, A.S., (2020). The Challenge of Research Supervision: The Experience of Lecturers in Various Academic Disciplines. 6th International Conference on Higher Education Advances (HEAd’20). Valencia, 2-5 June 2020. http://dx.doi.org/10.4995/HEAd20.2020.11234.

Wright, A. (2024). Postgraduate Supervision in a ChatGPT World: What’s Next? In: 10th International Conference on Higher Education Advances (HEAd’24). Valencia, 18-21 June 2024. https://doi.org/10.4995/HEAd24.2024.17244.

Wright, A. (2025a). Generative AI and Robust Educational Practices for Tourism, paper presented at the ATLAS Annual Conference, Universitat Rovira i Virgili, Vila-Seca, June 17-20, available at https://atlas-euro.org/wp-content/uploads/2025/06/ABSTRACT-BOOK-ATLAS-2025-Vila-seca-v2.pdf

Wright, A. (2025b). The GenAI Confusion for Educators: Where is the Light? Inclusion, Equity, and Generative AI in Education, paper presented at the AI&ED DEFI Cambridge University Conference, University of Cambridge, October 27-29.