Authors: Michelle Lai, Luke Dickens, Bonnie Buyuklieva, Janina Dewitz, Karen Stepanyan

Introduction

Accommodating neurodiversity in higher education often requires reasonable adjustments that take the form of additional time for completing the work, alternative assessment or examination settings, and the use of assistive technology among other options. An alternative approach to accommodating neurodiversity is to ensure that the material given to all students is universally accessible. We here examine AI based tools for improving the language used in tests to better support those with neurodiverse characteristics.

There is a significant proportion of the student body with neurodiverse characteristics. The statistics reported by HESA on self-disclosed learning disabilities such as dyslexia, dyspraxia, or AD(H)D show an increase of 26% between 2014-15 and 2021-22 (Tang et al., 2024). The total number of students with such learning differences in 2022-23 stands at 140,745 corresponding to roughly 6.5% of the total student population (HESA, 2025). Moreover, others have argued that this number is likely to underestimate the issue (Devine, J., 2024).

Neurodiverse characteristics, such as those associated with AD(H)D, autism, and dyslexia, have been linked to learning challenges associated with differences in cognitive function including: executive function (Southon, C., 2022), test anxiety (Mahak, S., 2024), and sensory sensitivities resulting in susceptibility to distraction (Irvine, B. et al., 2024.). This had led to calls for a revised approach to learning and assessment to include considerations of neurodiversity, e.g. (Sewell, A. and Park, J., 2021). Beyond academia, industry recruitment practices have shifted towards specifically taking into account neurodiversity of graduates as a way of obtaining greater competitive advantage (Borrett, 2024). Many universities are now leaning into offering greater support. In practice this takes the form of technology-based interventions, comprehensive support programmes, and transition into university and then employment (McDowall & Kiseleva, 2024). As a result of these changes the education sector is undergoing a paradigm shift that moves away from the traditional teacher-centred education to inclusive environments in Higher Education (Tang, Griffiths & Welch, 2024).

One approach to making education environments more inclusive is applying Universal Design principles to teaching materials. The principles of Universal Design advocate for developing environments that can be accessed, understood, and used to the greatest extent possible by all people, regardless of their ability or disability. Designs suitable for all people suggest that learning environments should strive to be as inclusive as possible, benefiting all stakeholders to the greatest degree possible, and reducing specific adjustments as much as possible (UCL-ARENA, 2024; Center for Universal Design, n.d.). This framework prioritises making the language of assessments more ‘neurodiversity-friendly’. We specifically focus on exploring the relevance of generative AI approaches for reformulating questions that are identified as problematic for neurodiverse cohorts. This project was funded by the UCL Centre for Humanities Education, which enabled us to conduct a case-study that looked at current examples of assessment, more specifically a sample of Multiple Choice Questions (MCQs), and the potential for making these more accessible to all students. In this case study, we look at two AI products that market themselves as generating textual outputs in neurodiversity-friendly language. We investigate how their suggested changes compared with the original questions set by two lecturers of technical subjects at UCL. The study was carried out by a funded postgraduate student and supported by academic staff at the Department of Information Studies, UCL, who co-authored this report.

Can AI Tools Make Assessment More Accessible?

Informed by the current body of research and by surveying available AI tools for diverse classroom applications (see Appendix 1), we shortlisted three tools based around two AI products (see Table 1). We then assessed whether these tools could make multiple-choice and short-answer questions written for assessment in technical modules more neurodiversity-friendly.

Our focus was on the AI tools that purport to be useful for making reading and language easier for neurodivergent people. We adopted Hemingway Editor and Goblin Tools Formaliser in our study. The latter product was tested under two modes: ‘More Accessible’ and ‘More to the point’ (see Table 1).

AI Tool	AI Technology Type	Short Description
Hemingway Editor	Rule based readability & style analyser. Light AI proofreading.	Live colour‑coding of “hard‑to‑read” sentences, passive voice, over‑used adverbs. Grade‑level calculation. AI pass for spelling/grammar and optional paraphrases (paid tier only).
Goblin: More accessible	LLM‑powered text simplification.	Rewrites any passage into plainer language at a lower reading level while preserving meaning.
Goblin: To the point	LLM‑based concision / summarisation.	Strips filler, tangents and redundancies, returning essence‑only version of the text.

Table 1: AI tools examined in this research, which were selected from the list of tools with the greatest potential to support language processing needs of neurodiverse students. Note: the full list of AI Tools considered for examination in this research is presented in Appendix 1.

Evaluation and Results

We combined the questions into question sets based on the topic and word length to accommodate the limitations of the tools. There were 7 sets of questions:

• Web Tech. 1, comprises 7 theoretical and introductory multiple-choice questions (MCQs) on the topic of the Internet and Web technologies;
• Web Tech. 2, comprises 7 MCQs containing HTML code;
• Web Tech. 3, comprises 7 theoretical MCQs on accessibility;
• Stats. 1, comprises a single MCQ in Statistics, which included a longer scenario and a longer text for the MCQ options;
• Stats. 2, comprises 4 MCQs on fundamental concepts in Statistics;
• Stats. 3, comprises on 2 MCQs that required simple statistical calculations;
• Stats. 4, comprises 3 MCQs that can be attributed to set theory – a foundational area of Mathematics.

The question bank comprises 31 questions, and was taken from an introductory web technologies modules taught at an undergraduate level, and an introductory statistics module taught to masters students, both at UCL.

Set	Name	Count	Length	Hemingway Editor				Goblin: More accessible				Goblin: To the point
				Red	Orange	Green	No change	Red	Orange	Green	No change	Red	Orange	Green	No change
1	Web Tech. 1	7	12.7	1	4	2	0	1	0	4	2	0	0	2	5
2	Web Tech. 2	7	19.4	1	1	4	1	0	1	5	1	0	1	5	1
3	Web Tech. 3	7	28.4	2	0	4	0	0	2	4	0	1	2	4	0
4	Stats. 1	1	99	0	0	1	0	0	0	1	0	0	0	1	0
5	Stats. 2	4	45.8	1	2	1	0	0	3	1	0	0	0	4	0
6	Stats. 3	2	59.5	0	0	2	0	0	0	2	0	0	0	2	0
7	Stats. 4	3	45	1	1	1	0	0	1	2	0	0	0	3	0

Table 2: Summary of studied question sets, including the number of questions in each set (Count), the average length of text in words (Length), and frequency of accuracy flags (Red, Orange and Green) by adopted AI tool for each question set or frequency of output with no change.

We passed each of the questions into the chosen three AI tools. We, therefore, had upto 93 suggested revisions along with the original set of 31 questions. Table 2 shows the number of questions (count) and average word count (length) for each set, along with the accuracy flags discussed further down.

One indication of how straightforward some text is to read is captured by the FORCAST (FORmula for Forecasting Average Single-Syllable word Count) readability formula (Scott, n.d.). We calculated the FORCAST scores for the original questions as well as the questions modified by the AI tools (Figure 1). A higher FORCAST score indicates a greater difficulty of written text.

While FORCAST scores can be useful, it is not always possible to gauge whether the output is clearer or more concise. However, there is no widely accepted framework for such an assessment, and we had to develop our own method for doing this consistently. Our assessment of clarity and conciseness was inspired by the earlier work of Bischof & Eppler (2011). However, we adopted a single Likert scale score for human judgment of each aspect for simplicity.

1	Language used is unclear, inappropriate, grammatically incorrect, or awkward.
3	Appropriate language and sentence structure is used. Certain words or turn of phrase are potentially awkward or incorrect, but do not impact the overall readability and understanding.
5	Appropriate language and sentence structure is used. Exceptionally easy to understand with no errors.

Table 3: Likert scale, used in evaluating Clarity of Language.

1	Overly wordy and convoluted. Severely impacts / delays the reader’s understanding.
3	Uses words of an appropriate difficulty for the topic at hand, with a standard / acceptable number of words to convey the intended meaning.
5	Uses the simplest and least number of words needed to convey the intended meaning.

Table 4: Likert scale, used in evaluating Conciseness.

All original and AI-generated question sets were assessed by one of the researchers using these scores for clarity and conciseness. For our analysis, a researcher was asked to rate the question for clarity on the given Likert scale (see Table 3). Next a researcher was asked to rate each question for conciseness on a Likert scale (Table 4). Lower scores, therefore, indicate a lack of clarity or conciseness.

The results of the mean scores of the Clarity and Conciseness evaluations are presented for all seven sets of questions in Figure 2 and Figure 3 respectively.

Finally, for each question in the sets the original authors of the test questions were asked if the questions produced by the AI tools retained the intended meaning with three possible judgement flags: green – the question retained the original meaning, orange – there were some issues with the meaning, and red – the question was no longer accurate given the context of the question. A final flag, gray, was given to questions that were unchanged by the tool. As indicated above, Table 2 presents the accuracy flag counts of the evaluated tools for each set. For readability these data are also plotted in Figure 4. It is clear from these results that these tools can perform differently across different types of questions, and no one tool performed without issues. Hemingway performed particularly badly on theoretical questions (Sets 1 and 3). It is also clear that Goblin: To the point performed well on questions containing code (Set 2). Overall accuracy, regardless the types of questions, does appear to be higher for both Goblin tools. However, this result should be viewed with caution due to a very small sample of questions used in the study. Furthermore, it is also important to study individual questions in greater detail to try and elicit why some of the questions were “tricky” for the AI tools.

Preliminary Findings

The study highlighted that the adoption of Generative AI tools for making text accessible to neurodiverse audiences does not necessarily offer improvements. The FORCAST scores for some of the sets increased (see Sets 2 & 4), making the questions potentially more complex as a result of using AI tools. However, occasionally (see Sets 1, 2 and 6, Figure 1) the use of AI-generated content has offered improvements in FORCAST scores. In some cases, the original questions scored worse on clarity and conciseness than the AI-generated alternatives (see Set 4, Figure 2 & 3). Similarly, improvements in clarity were evident in some instances too (see Set 2, 4 & 5 in Figure 2). Most importantly, however, the noted improvements were rather small, with the range of variation in scores remaining minimal for all of the adopted metrics.

Goblin ‘more to the point’ had a higher conciseness score overall (Figure 3), but had a higher FORCAST score as well, meaning that a higher grade / education level was required to read it. The FORCAST formula uses the number of single-syllable words to calculate a score, which may be why a more concise output using less ‘extra’ words will generate a higher FORCAST score.

Tool Specific Findings

Hemingway Editor

Hemingway Editor is a tool that offers readability statistics with the premium membership, including AI rewrite suggestions. This AI tool uses popular AI services, namely OpenAI, Anthropic, and Together. The interface of the Hemingway editor is easy to navigate, and is customisable. When feeding the questions in, the system recognised that the text was meant for a university-level reader and adjusted its suggestions accordingly. However, questions that contained HTML/CSS code were not processed correctly, failing to escape the tags, and changing the meaning of the question as a result.

Goblin Tools Formaliser: More Accessible

Goblin tools are a set of single-task tools that aim to break down and simplify tasks for neurodivergent people, and use “models from different providers, both open and closed source.” (goblin-tools, n.d.). The Formaliser tool converts texts with up to 15 prompts, such as ‘more accessible’ and ‘more to the point (unwaffle)’. There is a ‘spiciness level’ which controls how strongly the generated text will come across; for this study, it was set to the lowest level.

The ‘more accessible’ option often changed words to more commonly used and ‘lower grade’ words, as well as changing sentence structures. While this generally improved the readability and clarity of the outputs, the accuracy was sometimes affected, given the highly technical language of many questions.

Goblin Tools Formaliser: More to the Point

The ‘straight to the point’ Goblin Tools option often made little to no changes to the shorter questions, which suggests that in most cases the original question author had set questions using concise language. There tended to be no changes in wording, but this meant that the accuracy was usually unaffected, while the clarity of the question did not improve. For longer questions, there was a greater change in wording, which resulted in greater variation in clarity, and higher conciseness scores.

In addition, while the use of generative AI tools did not offer consistent performance in improving the FORCAST, clarity and conciseness scores, the review of generated alternative wording suggested by the tool were at times viewed as useful for formulating alternative ways of phrasing the questions.

Additional Reflections and Summary

The literature often refers to the clarity and conciseness as factors that could affect a neurodivergent student’s understanding of a question, relating to processing speed and cognitive load (e.g. Fakhoury et al., 2018).

In summary, our study suggests that generative AI tools can be useful as scaffolding tools for inclusive assessment design, when paired with informed human judgement. The Hemingway and Goblin Tools occasionally improved clarity, conciseness or readability, but introduced new difficulties related to the use of code snippets or altering meaning. It appears that generative AI can support producing neurodiversity‑friendly assessments, but it should only be adopted in an assistive role.

***

Dr Bonnie Boyana Buyuklieva FHEA, FRGS is a Lecturer (research) in Data Science for Society at UCL’s Department of Information Studies. Bonnie’s background is in architecture and computation (Foster and Partners; Bauhaus University Weimar), and she holds a PhD from the Bartlett Centre for Advanced Spatial Analysis (CASA, UCL).

Janina Dewitz has worked for over a decade as an Innovations Officer at UCL, having previously worked as a Learning Technologist and Learning Facilitator at Barking and Dagenham College. Janina earned a Bachelor’s in Humanities from the University of Hertfordshire in 1998 and a Higher National Certificate in Performing Arts from Barking & Dagenham College in 2004.

Dr Luke Dickens is an Associate Professor at UCL’s Department of Information Studies. Luke is also a Founding member and co-lead of the Knowledge Information and Data Science (KIDS) research group at UCL, and a Founding member of the cross-institutional Structured and Probabilistic Intelligent Knowledge Engineering (SPIKE) research group based at Imperial College.

Michelle Lai recently completed a Masters of Arts in Special and Inclusive Education (Autism) at UCL, having previously earned a BSc in Psychology with Education, Educational Psychology. Michelle is interested in how psychology can inform and enhance inclusive practices in special education and currently works as a Specialist Teaching Assistant at Ambitious about Autism.

Dr Karen Stepanyan is an Associate Professor (Teaching) in Computing and Information Systems. Based at the Department of Information Studies, he is leading the delivery of the Web Technologies (INST0007) module and Database Systems (INST0001).  He contributed to the development of the BSc Information in Society programme at UCL East. His research spans the inter-relation of information technologies and the concepts of knowledge.

***

References:

UCL-ARENA. (2024). “Inclusive education: Get started by making small changes to your education practice.” Teaching and Learning Retrieved April 1 2025, from https://www.ucl.ac.uk/teaching-learning/news/2024/nov/inclusive-education-get-started-making-small-changes-your-education-practice.

Bischof, N., & Eppler, M. J. (2011). Caring for Clarity in Knowledge Communication. J. Univers. Comput. Sci., 17(10), 1455-1473.

Borrett, A. (2024, Dec. 19) ‘UK employers eye “competitive advantage” in hiring neurodivergent workers’, Financial Times, accessed July 17 2025, Available at: https://www.ft.com/content/e692c571-b56b-425a-a7a0-3d8ae617080b.

Center for Universal Design. (n.d.). College of Design. https://design.ncsu.edu/research/center-for-universal-design/. Retrieved April 8, 2025 from https://design.ncsu.edu/research/center-for-universal-design/.

Devine, J. (2024). Comment: Neurodiversity and belongingness. Buckinghamshire New University. Retrieved July 18, 2025, from https://www.bucks.ac.uk/news/comment-neurodiversity-and-belongingness

Fakhoury, S., Ma, Y., Arnaoudova, V., & Adesope, O. (2018, May). The effect of poor source code lexicon and readability on developers’ cognitive load. In Proceedings of the 26th conference on program comprehension (pp. 286-296).

goblin-tools. About (n.d.). Retrieved July 18, 2025, from https://goblin.tools/About

HESA (Higher Education Statistics Agency). (2025, April 3). Who’s studying in HE?: Personal characteristics. Retrieved July 18, 2025, from https://www.hesa.ac.uk/data-and-analysis/students/whos-in-he/characteristics#breakdown

McDowall, A., & Kiseleva, M. (2024). A rapid review of supports for neurodivergent students in higher education. Implications for research and practice. Neurodiversity, 2. https://doi.org/10.1177/27546330241291769 (Original work published 2024)

Scott, B. (n.d.). Readability Scoring System PLUS. https://readabilityformulas.com/readability-scoring-system.php.

Tang, E. S. Y., Griffiths, A., & Welch, G. F. (2024). The Impact of Three Key Paradigm Shifts on Disability, Inclusion, and Autism in Higher Education in England: An Integrative Review. Trends in Higher Education, 3(1), 122-141. https://doi.org/10.3390/higheredu3010007

***

Appendix 1:

List of AI Tools considered for this study and classified by category (i.e. EF – Executive Function (planning, prioritizing, time management, task initiation); WM – Working Memory / information overload; LP – Language Processing / writing / reading load; SN – Sensory / modality flexibility (visual, auditory, text, speech); ANX – Reducing Anxiety (performance, math, participation); SOC – Social Communication load (participate without speaking live, review later); MATH – Math Conceptual/step scaffolding; ACC – General accessibility (alt input, captions, screen-reader friendly, etc.))

Tool	Short description	AI Tool Categories
Conker.ai	One‑click generator of differentiated quizzes & question banks	ANX, WM
Consensus AI	Answers research questions by ranking peer‑reviewed evidence	WM
Desmos	Web graphing calculator with interactive sliders & tables	MATH, SN
DreamBox	Adaptive K‑8 maths lessons that adjust every 60 seconds	MATH, ANX
Elicit AI	Semantic‑searches papers and auto‑extracts key findings	WM
Explain Paper	Click‑highlight any PDF sentence to get plain‑English explainer	LP
fireflies.ai	Live meeting recorder that transcribes, timestamps and summarises team calls	WM, SOC, ACC
GeoGebra	Dynamic geometry & graphing suite—visual proofs, 3‑D, AR	MATH, SN
goblin.tools Chef	Generates grocery lists & cooking steps from meal ideas	EF
goblin.tools Compiler	Condenses chat or notes into tidy bullet points	WM
goblin.tools Estimator	Predicts how long a task list will really take	EF
goblin.tools Formaliser	Rewrites text into more formal / academic register	LP
goblin.tools Judge	Rates whether your instructions are “clear enough”	EF
goblin.tools MagicToDo	Turns a vague task into an ordered, timed action plan	EF
goblin.tools Professor	Explains any concept at a chosen complexity level	WM
Grammarly	Real‑time grammar, tone and clarity checker inside browsers & docs	LP
Hemingway Editor	A readability tool that color‑codes dense or passive sentences, flags adverbs, and shows grade level so writers can simplify and clarify their prose.	LP
Heuristica	Generates interactive concept maps you can chat with	WM
IXL	Skills‑driven practice that levels up as students gain accuracy	MATH
Julius AI	Chat‑style data analyst that cleans, queries and plots spreadsheets	WM
Knewton Alta	Mastery‑based, adaptive courseware for math & science in HE	MATH
Litmaps	Builds visual citation maps to reveal research connections	WM
MathGPTPro / Mathos AI	LLM tutor that OCR‑reads handwritten maths and explains steps	MATH
MATHia	Carnegie Learning’s AI tutor that coaches each step of algebra problems	MATH, ANX
MathPapa	Symbolic algebra solver that shows stepwise solutions	MATH
Motion	AI calendar that auto‑schedules tasks against deadlines and reshuffles as priorities change	EF
Nuance Dragon Speech	High‑accuracy speech‑to‑text dictation across OS‑level apps	ACC, LP
Otter.ai	Live captioning & searchable transcripts for meetings and lectures	WM, SOC, ACC
PhET Interactive Simulations	Free, click‑and‑drag science & maths models (HTML5)	SN
SciSpace	All‑in‑one platform to search 200 M papers and ask PDFs questions	WM
Scite.AI	Shows whether later studies support or contradict a cited paper	WM
Scribe (To‑do)	Breaks complex goals into step‑by‑step checklists and sets reminders	EF
Scribe (Writing helper)	Generates SOPs and how‑to docs from screen recordings and prompts	LP, WM
Semantic Scholar (TLDR)	Gives one‑sentence abstract of any research paper	WM
SmartSparrow	Learner‑authored adaptive modules with branching feedback	WM
Socratic (Google)	Mobile camera‑based homework helper with brief video explainers	MATH, WM
Symbolab	Multi‑step calculator covering calculus, series, matrices, proof	MATH
Synthesia	Turns typed scripts into captioned avatar videos in 120+ languages	SN, ACC
Tavily	Real‑time search API built for LLMs & agents, returns source snippets	WM
Topmarks	Curated hub of short interactive literacy & numeracy games	SN, MATH
TXYZ.ai	AI “research OS” that finds, organises and chats over papers	WM
Unity ML‑Agents Toolkit	Open‑source SDK for building reinforcement‑learning–driven 3‑D sims and games	SN
Writer	Enterprise‑grade generative writing assistant with style‑guide enforcement	LP

 

In May 2024, an email popped into my inbox that would unexpectedly lead to a great academic experience: presenting at a national conference on student engagement. In the email from Abbi Shaw and Jesper Hansen, they were looking for students to take part in their research about how Arts and Humanities students engage with and think about AI. As an undergraduate Arts and Sciences student with an interest in artificial intelligence, faculty research being conducted about our perceptions of AI as students intrigued me, so I expressed my interest in being a part of this research. Fast forward to September, I was a student co-presenter at the 2024 RAISE (Researching, Advancing & Inspiring Student Engagement) Conference at the University of Leicester.

After expressing my interest in participating in the research, we were tasked with writing a series of reflections on five key questions and then joining a focus group to discuss them with other students. The questions were not just technical—they asked us to reflect on our personal relationships with AI, from the role it plays in our academic work to how we see it shaping our future careers. The questions were designed to enable us to dig deep into our feelings about generative AI, with questions about what reflections the term ‘generative AI’ triggers, or about how AI shows up in our daily life, or even in our studies at UCL. The introspective nature of the questions made me want to answer them and hear about what the other participants said.

A few months later, I received an exciting email from Abbi and Jesper about another opportunity: to be a student co-presenter at the RAISE Conference at the University of Leicester. If you have never heard of RAISE, it is a conference that focuses on research around student engagement, with themes like accessibility, equity, and celebrating diversity. This year’s theme was “Equity in Attainment & Student Success,” which was very fitting with the nature of our participation in this research. The purpose was to amplify student voices and make them part of the broader academic conversation. At the time, I did not know what the conference was about or what to expect from presenting, but I wanted to give it a go. So, Elisa Valentin (who has also written a blog on her experiences) and I joined the team as student co-presenters and had the chance to extend our participation.

Our presentation at RAISE focused on UCL’s ongoing efforts to incorporate a wide range of perspectives—from both staff and students—into the research process. Elisa and I shared our experiences of participating in faculty-led research, talking about how meaningful it was to be seen not just as participants, but as active contributors to the project.

Standing up there as a co-presenter, I realised that this kind of student engagement—where our insights are valued on an equal footing with those of staff—is exactly what conferences like RAISE are all about, and it fits into this broader discussion of co-creation. Co-creation was indeed a recurring theme throughout the conference, and it became clear that involving students in shaping their own learning is not just beneficial—it is necessary. Whether it is redesigning assessments, building inclusive learning environments, or making decisions about course content, there was a strong push for universities to break down traditional hierarchies and engage students as equal partners.

At the conference, there were people from various universities, all with different experiences and backgrounds, who gathered to discuss how we can reshape higher education. One of the parallel sessions that stood out to me was the one about “Why are students not attending in-person classes post-COVID-19?” by Conor Naughton (Education and Student Experience Manager at the University of Nottingham), Tom Lowe (Assistant Head of School (Student Experience), School of Finance and Accounting at the University of Westminster), and Tania Struetzel (Director of Student Success at Southampton Solent University). This session, delivered as an interactive workshop, explored the students’ perceptions of the necessity of in-person attendance after the pandemic. The majority of the room was academic staff members, and as a student, it was interesting to hear the academic staff discuss what they think the reasons for students’ low attendance to in-person classes might be.

As I listened to the discussions on post-COVID attendance, I realized just how complex student engagement is. While many staff members pointed to flexibility and convenience as factors, I found myself thinking about the importance of mental health, diverse learning preferences, and the need for universities to adapt. It was reassuring to see that the conversation was not about ‘blaming’ students for lower attendance, but about rethinking how we deliver education in a way that truly meets the needs of today’s learners.

This whole experience left me with a deeper understanding of equity in education, and how essential it is for students to be included in the shaping of our academic environments. It was pleasant to see so many educators and students working together to reimagine what inclusive education looks like. From being involved in the research to presenting at the RAISE Conference, this experience has shown me that students have a pivotal role to play in shaping not only our own learning journeys but also the broader academic landscape.

Author Bio: Selin Abdik is a second-year BASc Arts and Sciences student at UCL, specialising in the interdisciplinary application of technology. Selin has a strong interest in how technology can drive innovation and create impactful change across various fields. As a co-presenter at the 2024 RAISE Conference, Selin contributed insights on student involvement in research and policy changes within higher education. You can find out more about Selin’s work via Linkedin.

This project was supported by funding from UCL’s Centre for Humanities Education. The author wished to express their thanks to UCL CHE and the Randolph Quirk Endowment.

RAISE 2025 will be happening on September 4-5 September at the University of Glasgow.

The Centre for Humanities Education is pleased to offer up to £1000 for projects supporting educational research or enhancement of any kind. We particularly encourage applications in the areas of assessment and feedback as well as on AI, technology, and education (including exploratory projects). We also encourage collaborative projects that involve colleagues and students across career stages and disciplines.

Some ideas for what these funds could be used for:

• Events/conferences/workshops;
• To pay PGTAs or other students to undertake tasks relating to an Education-focused event/conference/activity (at the relevant standard UCL rate);
• To pay the expenses of visitors (travel/overnight hotel if necessary) who give talks (note that UCL Expenses Policy will apply to these);
• To run a small research project with student researchers
• To purchase resources/consumables.

What can they NOT be used for?

• They cannot be used for staff buy out;
• They cannot be used for the purchase of equipment. 

See the full list of previously funded projects here. 

The deadline for applications is Monday 22 April 2024 at 5pm. Projects need to be completed by 15 December 2024. 

Please click here to access the short online form to submit your application.

Good luck! We’re excited to see what comes next for CHE fund awardees!  

by Dr Simon Rowberry (UCL Centre for Publishing)

Edited 16 May 2024 to add: Inksights is now live! Check out the podcast here. 

In an era where digital landscapes evolve at lightning speed, equipping students with future-proof skills is more crucial than ever. With generative AI and multimedia platforms redefining professional boundaries, how can educational institutions stay ahead?

The answer: overhauling our curriculum for agility by embedding digital literacy throughout, and offering opportunities for students to work on real-world, career-enhancing projects.

A team of seven MA Publishing students, led by Amy, Lily and Mira, have been doing just this by working on Inksights, a podcast aimed at combining academic and industry perspectives on current publishing topics.

Each episode will blend interviews, reviews, debates, and news stories. Each member of team focuses on a different part of the podcast’s creation, production and marketing.

Finding your voice and preferred ways of collaborating

During the Inksights podcast’s pilot recording, the team tackled the art of collaboration head-on. The hosts encountered a challenge—avoiding interruptions—but ingeniously resolved it by adopting a lip balm stick as their “talking stick.” This simple yet effective strategy ensured clear, orderly discussions.

The Inksights team also skilfully navigated the challenge of sounding distinct and dynamic, steering clear of a “robotic” tone. The pilot showcases this success, achieved through planning and ensuring each host’s voice was unique and engaging. You can listen to it below (13:09 minutes long):

 

Methods of collaboration and working with guests were also challenges. While exploring various platforms, students discovered that each had its pros and cons. Zoom emerged as a standout for virtual sessions: the team praised it “for virtual interview/podcast recordings. The recording was quickly processed and available to access on the cloud by the production team.”

Working collaboratively on a podcast requires digital collaboration tools, which have proliferated over the last five years. The Inksights team are still finding the collaborative workflow that is most suitable for their needs. They are spending time to get this right early to avoid having to pivot to another setup later.

Our next steps include continuing to develop networks with similar groups of students and academics working on generative AI/podcasting across the University and the sector to gain further valuable insights and share good practice.

What Generative AI (currently) can and cannot do

Eden leads the generative AI project, focusing on images, videos, and animations. These are under-developed areas within publishing and require tools beyond Open AI’s Chat-GPT.

Through this scoping work, Eden notes: “I have become much more knowledgeable with this cutting-edge technology than industry professionals, putting me in a better position for job prospects.”

The projects highlight the constraints of today’s Generative AI tools in publishing. For example, it is possible to ask Chat-GPT to generate a book cover image, but this image appears in a square frame that requires further editing.

The file format of the output frequently clashes with current publishing workflows. This shifts human effort towards tedious tasks like cropping and format conversion, detracting from the creative focus on aesthetics and composition.

Likewise, most Generative AI services focus on a single media output, making cross-platform integrating challenging.

The Inksight podcast team’s experience with Zoom illustrates the benefits of human intervention as it “allows us to have video material to use for promotional purposes (Reels, YouTube, UCL publishing Instagram).”

This underscores the ongoing need to balance automation with human creativity, a vital skill for enhancing our students’ future employability.

The sustainability of extra-curricular media projects

The one-year format of our MA programme presents a significant hurdle for extracurricular activities as demonstrated by previous projects including Uncovering Publishing and Shelf Healing. Just as students hone their skills, they are ready to graduate, often having secured industry positions.

We do not yet have a practical solution to this challenge, but as the two projects continue to develop, we will look for ways to ensure greater continuity between cohorts and ensure teaching colleagues are equipped with the relevant skills to assist new groups of students.

The Inksights team are currently editing the first full episode of the podcast with the aim to release it by the end of Term 2.

Acknowledgements: Thanks to Amy, Eden, Lily and Mira for providing their thoughts on their progress to date. Simon used GPT-4 to improve the readability and conciseness of the first draft.