How Reverse Adaptation, Transmedia Storytelling and AI Tools Can Support Creative Independence and Artist Autonomy

Nadia Gasper, in her part-time, practice-based doctoral research, examines how novelisation and reverse adaptation can strengthen artist autonomy within the creative industries. The project explores how AI can support writers in research, editing, self-publishing, and promotion, while maintaining creative independence.

Nadia achieves this in multiple capacities: Firstly, AI is integrated into her structured methodology to organise and document creative work, manage research materials, and generate visual metrics to evaluate how theoretical approaches shape creative outcomes. In addition to supporting analytical reflection, Nadia also considers how AI tools can assist independent writers in producing promotional materials. She believes that AI can be used as a supplementary tool, one that enhances an artist’s capacity to produce literature and accompanying content such as book trailers and other promotional assets, especially for those without publishing budgets, while ensuring that an artist’s voice, originality, and creative agency remain firmly at the centre.

Finally, many publishers will often meet a quota for the demographics they hire, such as only have a limited number of authors focus on a minority demographic. Once this quota is achieved, publishers will not take on more writers focused on that topic of writing. By using AI to market their materials, Nadia mentioned that some authors would not need to go the traditional publishing route, bypassing industry gatekeeping. She stated: “Using AI responsibly allows me to be more than the demographic I’m supposed to appeal to. I can be a writer first.” Therefore, by situating AI within creative practice rather than treating it as a replacement for artistic labour, Nadia’s work demonstrates how emerging technologies can expand access and creative agency while preserving authorship and originality.

Donald Trump and Social Media AI

Jackson van Uden’s doctoral research examines the authoritarianisation of the United States Executive under Donald Trump, focusing on the administration’s use of social media and artificial intelligence to construct and amplify a contemporary “cult of personality.” The project analyses how AI-generated imagery has been used to shape executive identity and political messaging in new and highly visible ways.

His research demonstrates how AI models have produced hyper-masculine and nationalist depictions of Trump, which portray him as militaristic, divinely chosen, or positioned as a national saviour. These images then circulate rapidly across social media and official White House channels. These images often openly signal their AIgenerated nature, using stylised political “imaginaries” to reinforce narratives of strength and authority. At the same time, AI-generated derogatory images and videos targeting political opponents form part of a broader strategy of digital amplification and symbolic attack.

Situating these developments within executive studies and cult-of-personality scholarship, the project argues that AI represents a shift in how political authority isconstructed and disseminated. Unlike earlier forms of propaganda, such as that seen in Nazi Germany, AI-driven content relies on speed, virality, and shareability, enabling harmful narratives to be reproduced and amplified at scale. This student's research shows how AI reshapes how some politicians are approaching how they share their identity with the public. This has potentially far-reaching impacts for governments globally, where harmful narratives can easily be disseminated, impacting public belief, perception, and long-term political trajectories.

AI Representations of Muslim Women in the UK

Ruby Bashir’s doctoral research examines how AI operates within the broader social and institutional infrastructures shaping governance and security in the UnitedKingdom. Focusing on counterterrorism, securitisation, and national identity, the project analyses how algorithmic and data-driven systems are used to assess risk, monitor populations, and inform decision-making processes.

The research explores how these technologies influence the operationalisation of suspicion, belonging, and citizenship, often reinforcing existing racialised and gendered assumptions. By centring on the lived experiences of Muslim women, the project provides critical insight into how automated and semi-automated systems are encountered in everyday life, and how claims of technological objectivity intersect with power and marginalisation. As the researcher notes, “For Muslim women specifically, it matters because identity becomes publicly debated. AI-driven feeds can drive this and create environments [that are insecure for] already vulnerable populations. AI is indirectly directing the discourse.”

By situating AI within its social, political, and historical contexts, the project contributes to wider debates on ethical AI, bias, and accountability. Rather than treating artificial intelligence as a purely technical innovation, the research also demonstrates how algorithmic systems are embedded within governance structures and actively shape contemporary understandings of security and national identity.

Fine-tuned LLM for Annotation of Lobbying Records

Richard Wilkinson, a politics PhD student, is researching the perceptions of requests for administrative favours and fast-tracking by elected officials at the local level,focusing on those in Chile. The project draws on a large public dataset of approximately 175,000 lobbying records documenting meetings between citizens and mayors. Theserecords, written in Chilean Spanish and varying in length and detail, provide a richsource of text data for analysing how local oYicials respond to citizen requests.

Following an initial data exploration exercise using topic modelling and further manual classification of a subsample of the meeting data, Richard is now seeking to leverage afine-tuned large language model (LLM) to accurately annotate the remaining data inorder to examine the correlates of meeting subtypes across as many of Chile’s municipalities as possible. The project relies on a Spanish-language BERT model trained by the University of Chile’s Department of Computational Sciences.

In this context, AI functions as a tool to extract metrics from large amounts of text data to permit comparative analyses across a greater number of units.

How AI Learns to Understand Virtual Worlds

Yilin Yang’s research explores the evolving role of artificial intelligence as a system that not only performs tasks but increasingly interprets both virtual worlds and human lives.The project focuses on AI applications in the gaming industry and in personal development contexts, examining how AI systems model environments, behaviours,and decision-making processes.

Rather than approaching AI solely as a technical tool, the research conceptualises it as an interpretive and generative system that actively shapes both digital experiences and human self-understanding. It investigates how AI is used to construct and optimise virtual worlds — such as in game content generation and marketing — and how similar logics are applied to personal growth, including identifying individual strengths, imagining future trajectories, and translating aspirations into actionable pathways.

By situating these developments within broader questions of values, agency, and control, the project contributes to ongoing debates about the societal implicationsof AI. It highlights how AI systems increasingly participate in defining what is considered meaningful, successful, or desirable, and raises critical questions about whose values are embedded in these systems and how human autonomy can be maintained in AI mediated environments.

Digital Sustainability – Exploring the transformational capabilities through the lensof ESG Framework©

Gayathri Srinivasan Parthasarathy’s master’s research builds on her prior work on the topic of digital sustainability, examining the relationship between AI and sustainable development. The project explores how AI systems intersect with environmental, social, and governance framework, considering the potential benefits and broader systemic impacts. Her research investigates on how AI can be applied within sustainability focused initiatives, while also critically assessing the environmental costs and infrastructural demands associated with AI technologies themselves. This also includes the use of AI as a tool in diverse fields, examining the issues such as energy consumption and the long-term implications of embedding AI within the sustainability strategies. By situating AI within the wider field of digital sustainability, the research questions on how the use of AI must align with improved technology along with sustainable governance, ethical frameworks, and long-term societal goals.

OCR for Degraded Manuscripts Using AI

Audrey Jordan’s doctoral research investigates the use of AI for Handwritten Text Recognition (HTR) in degraded historical scripts, with a particular focus on data efficient approaches suitable for humanities collections. Audrey’s work recognises an ongoing problem with archival sources because they often contain limited training data and exhibit significant physical deterioration or script variation, making standard AI methods difficult to apply effectively. Further, manual transcription is highly timeintensiveand often restricts the scope of historical research. As the researcherexplains, “Transcribing historical things is a nightmare and [there is] too much [data].With low staff, people will not do it and that rules out history [as a subject of study].”

In response, the research develops data-efficient deep learning techniques tailored to low-resource archival environments. Using approximately 600 labelled manuscript samples, the project applies random augmentation to improve robustness and trains arecurrent neural network (RNN) on word-segmented medieval texts to predict sequential content across entire manuscripts. In experimental trials, the model has outperformed comparable commercial systems.

By adapting AI techniques, the research demonstrates how computational methods can expand access to archival materials while preserving the researcher’s role in interpretation and contextual analysis.

Generative AI vs Perceptual Hashes: Security and Privacy Risks for Image-BasedSexual Abuse Removal Tools

Sophie Hawkes PhD research examines the security and privacy risks associated with perceptual hashing systems used to detect and remove image-based sexual abuse material online. Perceptual hashes are designed to function as irreversible digital fingerprints, enabling platforms to identify harmful content without storing original images. However, the project critically evaluates the robustness of these systems in the context of advances in AI.

By training generative adversarial networks (GANs) on perceptual hash values derived from widely used systems such as Facebook’s PDQ and Apple’s Neural Hash, the research demonstrates that approximate reconstructions of original images can be generated, even using consumer-grade hardware. These findings challenge assumptions about the irreversibility and safety of perceptual hashing technologies and highlight the risk that sensitive image data could be exposed if hash values are compromised.

In response, the project proposes more secure, privacy-preserving alternatives, including private set intersection protocols that enable content matching withoutrevealing underlying image data. By combining adversarial testing with defensive design, the research contributes to broader debates on secure AI deployment, digital privacy, and the responsible governance of automated content moderation systems.

Investigating Techniques for Using GenAI in Code Refactoring and Translation

Nathan Rutherford’s PhD is exploring how AI can help update old software by translating programmes written in old code (C) into newer, safer code (Rust). Many organisations still depend on old code, which can be vulnerable to security problems and is expensive to rewrite manually, so using AI as an automated “translator'” can save significant time and money. His work shows that AI can often produce more natural,human-like translations than traditional tools, potentially improving security, readability, and scalability. However, there are drawbacks to using this approach, such as that AI can sometimes “hallucinate'' incorrect code or produce translations that technically run but don’t behave exactly like the original programme. To address this,the student suggests careful testing, having AI systems review each other’s outputs,and combining AI with established code-analysis tools. Overall, the research presents AI as a useful tool for updating legacy systems while realistically acknowledging that it must be carefully checked and supported by other tools to be trusted.

Beyond the Sensory Room: AI-Driven Curation of Neuro-inclusive Environments

Paulina Brosz’s PhD is interested in developing an AI-driven interactive system using augmented and virtual reality (AR/VR) hardware to explore the dynamic curation of sensory environments. The project investigates the feasibility of adapting visual and auditory stimuli in real time to support individuals who experience high levels of distraction or who have specific sensory needs. The research examines how artificial intelligence can analyse user input and environmental data to modify immersive spaces responsively, creating personalised sensory settings that adjust to cognitive or perceptual demands.

Situating this work within broader discussions of HCI and inclusive design, the research contributes to ongoing debates about how AI-enabled environments can support concentration, wellbeing, and user agency. Rather than treating immersive technologies as fixed experiences, the project demonstrates how AI can enable responsive systems that adapt to individual sensory profiles while maintaining user control and interpretive engagement.

Adaptive Reading with AI and Eye-Tracking

Manuel Muñoz’s PhD research addresses cognitive challenges associated with digital reading, including information overload and fragmented attention. His project centres around developing an AI-driven “cognitive partner” that uses real-time eye-tracking datato infer a reader’s cognitive state and provide tailored, transparent support. By analysing patterns in gaze behaviour, such as fixations or backward movements (regressions), the system can identify when a reader may be experiencing difficulty and respond by offering contextual assistance, such as definitions, translations, sentence rephrasing,or attention-redirection cues. In this way, the technology dynamically adapts to the reader’s needs as they engage with text.

The research also prioritises transparency and rapid personalisation to move beyond generic, "one-size-fits-all" reading aids. By integrating Explainable AI with Large Language Models, the system aims to build user trust and foster metacognition. Through a combination of empirical studies, the project aims to effectively support adiverse range of readers, including language learners or neurodivergent users.

By combining this adaptive framework with gaze tracking and cognitive theory, the project aims to demonstrate how responsive and personalised technologies can enhance accessibility and comprehension while maintaining user agency. It also highlights the potential of AI-driven systems to support inclusive digital reading environments.

What Can Large Language Models Tell Us About Human Reading?

Haibei Wang’s doctoral research examines how readers use prediction during reading and how advances in AI can improve theoretical models of reading behaviour. The ability to read effectively is closely linked to educational attainment, economic opportunity, health, and self-advocacy. Many influential theories of skilled reading propose that readers anticipate upcoming words to increase the speed and efficiency of comprehension. However, existing models, including widely used eye-movement models such as the E-Z Reader model, rely on a traditional measure known as cloze probability as a proxy for predictability. This metric, developed over seventy years ago, offers only a limited and indirect measure of how predictive processes operate.

The project leverages recent developments in large language models to derive more precise and theoretically grounded measures of word predictability. Using AI-generated predictability metrics across three large eye-movement corpora, the research investigates whether these measures better explain readers’ eye movements when compared to traditional approaches. The project also tests whether integrating AIderived metrics into computational models such as E-Z Reader improves their explanatory power.

In addition, the research includes new empirical studies examining how top-down predictive processes interact with bottom-up visual input during reading. By combining cognitive psychology with AI-based modelling, the project advances the understanding of how predictive mechanisms support fluent reading and contributes to more refined theories of human language processing.